CA2504567A1 - Method of color image processing to eliminate shadows and reflections - Google Patents
Method of color image processing to eliminate shadows and reflections Download PDFInfo
- Publication number
- CA2504567A1 CA2504567A1 CA002504567A CA2504567A CA2504567A1 CA 2504567 A1 CA2504567 A1 CA 2504567A1 CA 002504567 A CA002504567 A CA 002504567A CA 2504567 A CA2504567 A CA 2504567A CA 2504567 A1 CA2504567 A1 CA 2504567A1
- Authority
- CA
- Canada
- Prior art keywords
- image
- component
- components
- color image
- image processing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 15
- 230000009466 transformation Effects 0.000 claims description 25
- 238000001914 filtration Methods 0.000 claims description 2
- 230000011218 segmentation Effects 0.000 abstract description 6
- 230000003071 parasitic effect Effects 0.000 description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 239000003086 colorant Substances 0.000 description 2
- 238000003709 image segmentation Methods 0.000 description 2
- 238000010521 absorption reaction Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
Classifications
-
- G06T5/94—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/40—Image enhancement or restoration by the use of histogram techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30248—Vehicle exterior or interior
- G06T2207/30252—Vehicle exterior; Vicinity of vehicle
- G06T2207/30256—Lane; Road marking
Abstract
A method of color image processing comprises at least the following steps: t he color image is converted into an intermediate image comprising components th at depend solely on the H and S components of the original image in an HSV or H LS representation; the intermediate image comprising at least two components X and Y, determined by functions especially of the H component, these function s taking the same value when the H component is zero or equal to one; a new image comprising only one component is generated from the intermediate image , the component of this new image being a function of the components of the intermediate image. The disclosed method can be applied especially to the heightening of contrast in remote surveillance, and to image processing befo re segmentation in mobile robotics.
Description
Method of color image processing to eliminate shadows and reflections BACKGROUND OF THE INVENTION
The present invention relates to a method for the processing of color images to eliminate shadows. It can be applied especially to the heightening of contrast in remote surveillance, and to image processing before segmentation in mobile robotics.
In applications of mobile robotics, operations of image segmentation are carried out. Segmentation makes it possible to detect and localize the edges of roads for example. The shadows of trees on these roads or reflections from puddles generate parasitic segments. These shadows and reflections constitute parasitic information. The invention makes it possible to eliminate this parasitic information by carrying out a processing of color images that are to be segmented.
SUMMARY.OF THE INVENTION
An object of the invention is a method of color image processing, comprising the following steps:
- the color image is converted into an intermediate image comprising components that depend solely on the H and S components of the original image in an HSV or HLS representation;
- the intermediate image comprising at least two components X and Y, 2o determined by functions especially of the H component, these functions taking the same value when the H component is zero or equal to one;
- a new image comprising only one component is generated from the intermediate image, the component of this new image being a function of the components of the intermediate image.
The method according to the invention has many advantages:
- it can be used to extract the useful information from a color image with a minimum loss of information;
- it can be used to compress the information in a monochromatic image;
3o - it is totally non-parametrical and can therefore be applied to all images with the same efficiency, without any particular setting, whatever the scene.
After a processing operation of this kind, it is possible to use monochromatic image segmentation algorithms while at the same time exploiting the wealth of information of a color image in getting rid of errors due to shadows or reflections of water films.
BRIEF DESCRIPTION OF THE DRAWINGS
The invention shall now be described in great deal in the context of a particular example of a practical implementation. In this description, reference shall be made to the figures of the appended drawings, of which:
- Figure 1 represents the RGB color space;
- Figure 2 represents the HSV color space;
- Figure 3 is a graph showing the steps of implementation of the 1o invention according to an advantageous mode using texture attributes;
- Figure 4 exemplifies a result of a Karhunen-Loeve transformation operation;
- Figure 5 shows an alternative implementation of the example shown in figure 3;
~5 - Figure 6 is an exemplary histogram of a new image obtained by the implementation of the invention to which a filtering of pixels and an adjustment of the dynamic range is applied;
- Figure 7 shows another exemplary adjustment of the dynamic range.
MORE DETAILED DESCRIPTION
2o Referring to figure 1, a color image is generally described in the RGB (red, green, blue) space. The color of each pixel is represented therein by three components R (red), G (green), B (blue). These components may be represented on three axes 11, 12, 13 in Cartesian coordinates. The primary colors, cyan, blue, magenta, red, yellow and green, form the corners 25 of a cube 10 in this space. Black and white form opposite corners of this cube. The segment connecting these points is a main diagonal of this cube on which only the luminance L varies.
Referring now to figure 2, other color spaces exist. In the HSV
(hue, saturation, value) space, color is represented by three components H, 3o S, and V. These components may be represented in a referential system of cylindrical coordinates. The H component is represented by an angle 21.
The S and V components are represented on two axes 22 and 23. The H
component is the hue. This component makes it possible for example to distinguish red and yellow. The S component is saturation, namely the purity 35 of the color. This component makes it possible for example to distinguish red and pink. The V component is the value or intensity, namely the quantity of light. This component makes it possible, for example, to distinguish between dark gray and light gray, or again between dark red and light red. It is assumed hereinafter in the description that the components H, S, V are normalized between 0 and 1. The following explanations may be transposed in an obvious way if these components are not normalized between 0 and 1.
The passage from the RGB representation to the HSV
representation is done according to known analytical formulae. The primary colors and black form a six-faced pseudo-cone (hexcone) 20. The plane with ~o a constant value of V corresponds to the projection of the RGB cube parallel to its main diagonal.
The Applicant has noted that the value or intensity V does not carry a great deal of useful information because the elements of the image to be segmented are distinguished far more by their spectral properties than by their reflection or their total absorption. Furthermore, the value or intensity V
carries information that leads the segmentation algorithms into error, since shadowed and non-shadowed zones of a same object (a road for example) have very different intensities (with variation of intensity at the source).
Similarly, a water film on the road gives rise to a strong reflection and, 2o therefore, a very different intensity from that of a dry or simply wet portion on the same road (with variation of intensity by modification of the properties of reflection).
In order to eliminate the information on value or intensity, the color image is converted into an intermediate image in which the V component is absent. This intermediate image includes components that depend solely on the H and S components of the original image in an HSV representation.
This makes it possible to eliminate cast shadows and reflections due to water films. This furthermore has the advantage of reducing the quantity of information present in the image (by the elimination of a 3o component).
The conversion from the RGB space to the HSV space gives rise to singularities of hue for certain points. Thus, for the achromatic points (R=G=B), a small variation in RGB may lead to a big variation in H.
Furthermore, for the points in the vicinity of the axis of origin of H (G=B, R>(G+B)l2), a small disturbance in the components R, G, B may lead to major variations in H (a change in the sign of H).
Furthermore, the distribution of the converted values is not uniform. Digitization may lead to peaks and holes in the distribution of the values. In other words, the information is noise-ridden around V=0.
Advantageously, components X and Y are used in the intermediate image. These components are determined by functions especially of the H component, and take the same value when the H
component is zero or equal to one. Thus, the similarities induced by the H
1o component, proper to the system of cylindrical coordinates, are eliminated.
These functions are written as GY (H~ and GY (H~. They therefore verify the following relationships: GX (0) = GX ~1~ and G,, (0~= GY (1~. The components X and Y are then determined by:
X = G~ ~H~
Y=GY~H
For example, the following functions may be used:
G.~. ~H~ = cos~2TCH - ~p~
,.. GY {H~ = sin~2~-H - ~p~
where ~p is a constant.
2o A third component can be added to this intermediate image:
saturation S. Thus, in the intermediate image, there is hue information H and saturation information S, while at the same time the singularities have been eliminated.
Advantageously, it is possible to combine hue information H and saturation information S in only two components of the intermediate image.
To this end, the components X and Y are determined by functions not only of the H component but also of the S component. In the same way as above, these functions take the same value when the H component is zero or equal to one. Furthermore, these functions tend toward zero when the S
3o component tends toward zero.
These functions are written as Ga (H, S~ and G;, (H, S~ . They therefore verify the following relationships:
G.Y ~0, s~ = G.~. ~1, S
GY ~0, S) = G,, ~1, S
GX (H, S~ ~ 0 s -~ o GY(H,S) ~ 0 s ~ o For example, it is possible to use a product of two independent functions respectively of the S component and of the H component:
The present invention relates to a method for the processing of color images to eliminate shadows. It can be applied especially to the heightening of contrast in remote surveillance, and to image processing before segmentation in mobile robotics.
In applications of mobile robotics, operations of image segmentation are carried out. Segmentation makes it possible to detect and localize the edges of roads for example. The shadows of trees on these roads or reflections from puddles generate parasitic segments. These shadows and reflections constitute parasitic information. The invention makes it possible to eliminate this parasitic information by carrying out a processing of color images that are to be segmented.
SUMMARY.OF THE INVENTION
An object of the invention is a method of color image processing, comprising the following steps:
- the color image is converted into an intermediate image comprising components that depend solely on the H and S components of the original image in an HSV or HLS representation;
- the intermediate image comprising at least two components X and Y, 2o determined by functions especially of the H component, these functions taking the same value when the H component is zero or equal to one;
- a new image comprising only one component is generated from the intermediate image, the component of this new image being a function of the components of the intermediate image.
The method according to the invention has many advantages:
- it can be used to extract the useful information from a color image with a minimum loss of information;
- it can be used to compress the information in a monochromatic image;
3o - it is totally non-parametrical and can therefore be applied to all images with the same efficiency, without any particular setting, whatever the scene.
After a processing operation of this kind, it is possible to use monochromatic image segmentation algorithms while at the same time exploiting the wealth of information of a color image in getting rid of errors due to shadows or reflections of water films.
BRIEF DESCRIPTION OF THE DRAWINGS
The invention shall now be described in great deal in the context of a particular example of a practical implementation. In this description, reference shall be made to the figures of the appended drawings, of which:
- Figure 1 represents the RGB color space;
- Figure 2 represents the HSV color space;
- Figure 3 is a graph showing the steps of implementation of the 1o invention according to an advantageous mode using texture attributes;
- Figure 4 exemplifies a result of a Karhunen-Loeve transformation operation;
- Figure 5 shows an alternative implementation of the example shown in figure 3;
~5 - Figure 6 is an exemplary histogram of a new image obtained by the implementation of the invention to which a filtering of pixels and an adjustment of the dynamic range is applied;
- Figure 7 shows another exemplary adjustment of the dynamic range.
MORE DETAILED DESCRIPTION
2o Referring to figure 1, a color image is generally described in the RGB (red, green, blue) space. The color of each pixel is represented therein by three components R (red), G (green), B (blue). These components may be represented on three axes 11, 12, 13 in Cartesian coordinates. The primary colors, cyan, blue, magenta, red, yellow and green, form the corners 25 of a cube 10 in this space. Black and white form opposite corners of this cube. The segment connecting these points is a main diagonal of this cube on which only the luminance L varies.
Referring now to figure 2, other color spaces exist. In the HSV
(hue, saturation, value) space, color is represented by three components H, 3o S, and V. These components may be represented in a referential system of cylindrical coordinates. The H component is represented by an angle 21.
The S and V components are represented on two axes 22 and 23. The H
component is the hue. This component makes it possible for example to distinguish red and yellow. The S component is saturation, namely the purity 35 of the color. This component makes it possible for example to distinguish red and pink. The V component is the value or intensity, namely the quantity of light. This component makes it possible, for example, to distinguish between dark gray and light gray, or again between dark red and light red. It is assumed hereinafter in the description that the components H, S, V are normalized between 0 and 1. The following explanations may be transposed in an obvious way if these components are not normalized between 0 and 1.
The passage from the RGB representation to the HSV
representation is done according to known analytical formulae. The primary colors and black form a six-faced pseudo-cone (hexcone) 20. The plane with ~o a constant value of V corresponds to the projection of the RGB cube parallel to its main diagonal.
The Applicant has noted that the value or intensity V does not carry a great deal of useful information because the elements of the image to be segmented are distinguished far more by their spectral properties than by their reflection or their total absorption. Furthermore, the value or intensity V
carries information that leads the segmentation algorithms into error, since shadowed and non-shadowed zones of a same object (a road for example) have very different intensities (with variation of intensity at the source).
Similarly, a water film on the road gives rise to a strong reflection and, 2o therefore, a very different intensity from that of a dry or simply wet portion on the same road (with variation of intensity by modification of the properties of reflection).
In order to eliminate the information on value or intensity, the color image is converted into an intermediate image in which the V component is absent. This intermediate image includes components that depend solely on the H and S components of the original image in an HSV representation.
This makes it possible to eliminate cast shadows and reflections due to water films. This furthermore has the advantage of reducing the quantity of information present in the image (by the elimination of a 3o component).
The conversion from the RGB space to the HSV space gives rise to singularities of hue for certain points. Thus, for the achromatic points (R=G=B), a small variation in RGB may lead to a big variation in H.
Furthermore, for the points in the vicinity of the axis of origin of H (G=B, R>(G+B)l2), a small disturbance in the components R, G, B may lead to major variations in H (a change in the sign of H).
Furthermore, the distribution of the converted values is not uniform. Digitization may lead to peaks and holes in the distribution of the values. In other words, the information is noise-ridden around V=0.
Advantageously, components X and Y are used in the intermediate image. These components are determined by functions especially of the H component, and take the same value when the H
component is zero or equal to one. Thus, the similarities induced by the H
1o component, proper to the system of cylindrical coordinates, are eliminated.
These functions are written as GY (H~ and GY (H~. They therefore verify the following relationships: GX (0) = GX ~1~ and G,, (0~= GY (1~. The components X and Y are then determined by:
X = G~ ~H~
Y=GY~H
For example, the following functions may be used:
G.~. ~H~ = cos~2TCH - ~p~
,.. GY {H~ = sin~2~-H - ~p~
where ~p is a constant.
2o A third component can be added to this intermediate image:
saturation S. Thus, in the intermediate image, there is hue information H and saturation information S, while at the same time the singularities have been eliminated.
Advantageously, it is possible to combine hue information H and saturation information S in only two components of the intermediate image.
To this end, the components X and Y are determined by functions not only of the H component but also of the S component. In the same way as above, these functions take the same value when the H component is zero or equal to one. Furthermore, these functions tend toward zero when the S
3o component tends toward zero.
These functions are written as Ga (H, S~ and G;, (H, S~ . They therefore verify the following relationships:
G.Y ~0, s~ = G.~. ~1, S
GY ~0, S) = G,, ~1, S
GX (H, S~ ~ 0 s -~ o GY(H,S) ~ 0 s ~ o For example, it is possible to use a product of two independent functions respectively of the S component and of the H component:
5 GX(FI,S~=F'x('S~xG~(H~
GY (H, S> = FY ~S~ x GY (H~
The functions FX (S~, FY (S) of the S component are functions tending toward zero when the S component tends toward zero. These functions may be, for example, the identity function (for example FY (S~= S ) or the square function (for exampleFx (S~= SZ ). Preferably, monotonic and continuous functions will be used.
Referring now to figure 3, in an advantageous variant, an intermediate image 34 is generated. This intermediate image furthermore comprises texture attributes. The texture attributes are determined from an image 32 comprising the components X and Y for example, and not from the initial color image 30 encoded in the RGB space. Indeed, it is sought first of all to eliminate the shadows and the reflections.
For example the initial color image 30 is converted into an image 31 with two components, the H and S components. Then, the image 31 is 2o converted into an image 32 that comprises the two components X and Y
defined here above. For each pixel of the image 32, texture attributes are determined. These attributes may be, for example, the attributes defined by Haralick. Of course, other attributes may be used, through a fractal approach for example. An image 33 is obtained with texture attributes. The components X and Y and the texture attributes are combined to form the intermediate image 34, to which the remaining processing operations are applied. It is thus possible to apply a multitude of transformations to the intermediate image without being preoccupied by shadows and reflections.
In order to obtain a monochromatic image, a new image is 3o generated from the intermediate image obtained. In other words, a transformation is applied to the intermediate image so as to keep only one component. The single component of the new image is a function of the components of the intermediate image.
GY (H, S> = FY ~S~ x GY (H~
The functions FX (S~, FY (S) of the S component are functions tending toward zero when the S component tends toward zero. These functions may be, for example, the identity function (for example FY (S~= S ) or the square function (for exampleFx (S~= SZ ). Preferably, monotonic and continuous functions will be used.
Referring now to figure 3, in an advantageous variant, an intermediate image 34 is generated. This intermediate image furthermore comprises texture attributes. The texture attributes are determined from an image 32 comprising the components X and Y for example, and not from the initial color image 30 encoded in the RGB space. Indeed, it is sought first of all to eliminate the shadows and the reflections.
For example the initial color image 30 is converted into an image 31 with two components, the H and S components. Then, the image 31 is 2o converted into an image 32 that comprises the two components X and Y
defined here above. For each pixel of the image 32, texture attributes are determined. These attributes may be, for example, the attributes defined by Haralick. Of course, other attributes may be used, through a fractal approach for example. An image 33 is obtained with texture attributes. The components X and Y and the texture attributes are combined to form the intermediate image 34, to which the remaining processing operations are applied. It is thus possible to apply a multitude of transformations to the intermediate image without being preoccupied by shadows and reflections.
In order to obtain a monochromatic image, a new image is 3o generated from the intermediate image obtained. In other words, a transformation is applied to the intermediate image so as to keep only one component. The single component of the new image is a function of the components of the intermediate image.
Referring now to figure 4, in an advantageous embodiment of the invention, the new image is generated by keeping only the first component of the Karhunen-Loeve transformation.
This transformation is applied in the referential system D1, D2 of the eigen vectors are associated with the covariance matrix of the pixels of the intermediate image. There is thus a passage from the X, Y referential system to the D1, D2 referential system. As can be seen in the exemplary distribution of the pixels 40 of the image in these referential systems, this transformation enables the decorrelation of the information present in the ~o image. The first component D1 is the most discriminatory component, i.e. it is this component that maximises the information. This facilitates discrimination in the segmentation processing operations applied to the new image.
According to an alternative mode of implementation, a linear approximation of the Karhunen-Loeve transformation is applied, not the Karhunen-Loeve transformation itself. To this end, the Karhunen-Loeve transformation is applied to a representative set of images (for example eight images). The images of this set have characteristics similar to the images to which it is sought to apply the linear approximation. An average is then taken of the changes in referential system obtained by the exact Karhunen-Loeve transformation. A average change of referential system is obtained, and this referential will be used as an approximation of the Karhunen-Loeve transformation.
According to another mode of implementation, the new image is generated by projecting the components of the intermediate image in the plane in which the dynamic range or the mean standard deviation is the greatest.
Reference is now made to figure 5 which represents an alternative mode of implementation with respect to the example shown in figure 3. In 3o the same way, from the initial color image 50, an image with two components, the H and S components, is generated. Then an image 52 comprising the two components X and Y is generated. Then the Karhunen-Loeve transformation (or a linear approximation ,of this transformation) is applied to the components X and Y. The first component of the result of this transformation is kept to form an image 53 comprising a single component.
This transformation is applied in the referential system D1, D2 of the eigen vectors are associated with the covariance matrix of the pixels of the intermediate image. There is thus a passage from the X, Y referential system to the D1, D2 referential system. As can be seen in the exemplary distribution of the pixels 40 of the image in these referential systems, this transformation enables the decorrelation of the information present in the ~o image. The first component D1 is the most discriminatory component, i.e. it is this component that maximises the information. This facilitates discrimination in the segmentation processing operations applied to the new image.
According to an alternative mode of implementation, a linear approximation of the Karhunen-Loeve transformation is applied, not the Karhunen-Loeve transformation itself. To this end, the Karhunen-Loeve transformation is applied to a representative set of images (for example eight images). The images of this set have characteristics similar to the images to which it is sought to apply the linear approximation. An average is then taken of the changes in referential system obtained by the exact Karhunen-Loeve transformation. A average change of referential system is obtained, and this referential will be used as an approximation of the Karhunen-Loeve transformation.
According to another mode of implementation, the new image is generated by projecting the components of the intermediate image in the plane in which the dynamic range or the mean standard deviation is the greatest.
Reference is now made to figure 5 which represents an alternative mode of implementation with respect to the example shown in figure 3. In 3o the same way, from the initial color image 50, an image with two components, the H and S components, is generated. Then an image 52 comprising the two components X and Y is generated. Then the Karhunen-Loeve transformation (or a linear approximation ,of this transformation) is applied to the components X and Y. The first component of the result of this transformation is kept to form an image 53 comprising a single component.
Texture attributes of this image 53 are determined. These texture attributes form an image 54. The single component of the image 53 and the texture attributes (image 54) are combined to form the intermediate image 55 comprising color information and texture information. The remaining processing operations are applied to this intermediate image 55.
The Karhunen-Loeve transformation can thus be applied a second time, for example to obtain a monochromatic image.
Reference is now made to figure 6 which represents an exemplary histogram 60 of the component of the monochromatic image. The x-axis 61 ~o represents the values taken by the pixels. The y-axis 62 represents the number of pixels in the image having a determined value. A non-negligible range of the values is taken up by pixels that are very small in number at the two extreme ends 63a, 63b of the histogram.
Advantageously the lightest and the darkest pixels, which are present in very small numbers, are filtered. This makes it possible to avoid any loss of dynamic range through the occupying of a major part of the range of values by pixels that carry no additional information. In other words, the darkest and the lightest pixels, which represent a determined fraction of the total number of pixels of the image, are filtered. To this end, a number of 2o pixels to be eliminated from the image is determined. This number of pixels may be expressed as a percentage of the total number of pixels of the image.
A minimum value VMIN and a maximum value VMAX of the pixels are determined, so that the number of pixels of the image whose value is below VMIN or above VMAX is as close as possible to the number of pixels to be eliminated. For all the pixels whose value is beyond the values VMIN and VMAX, the value of the corresponding limit is assigned.
The values VMIN and VMAX may be determined iteratively.
During an iteration, a value 64 of the y-axis 62 is selected in going through the axis from the bottom to the top. The smallest value VMIN and the 3o greatest value VMAX of the y-axis 61, corresponding to the intersections 64a, 64b of the histogram 60 with the selected value 64, are determined. The operation is stopped when the number of pixels to be eliminated for the selected value is greater than the determined fraction of the total number of pixels of the image.
The Karhunen-Loeve transformation can thus be applied a second time, for example to obtain a monochromatic image.
Reference is now made to figure 6 which represents an exemplary histogram 60 of the component of the monochromatic image. The x-axis 61 ~o represents the values taken by the pixels. The y-axis 62 represents the number of pixels in the image having a determined value. A non-negligible range of the values is taken up by pixels that are very small in number at the two extreme ends 63a, 63b of the histogram.
Advantageously the lightest and the darkest pixels, which are present in very small numbers, are filtered. This makes it possible to avoid any loss of dynamic range through the occupying of a major part of the range of values by pixels that carry no additional information. In other words, the darkest and the lightest pixels, which represent a determined fraction of the total number of pixels of the image, are filtered. To this end, a number of 2o pixels to be eliminated from the image is determined. This number of pixels may be expressed as a percentage of the total number of pixels of the image.
A minimum value VMIN and a maximum value VMAX of the pixels are determined, so that the number of pixels of the image whose value is below VMIN or above VMAX is as close as possible to the number of pixels to be eliminated. For all the pixels whose value is beyond the values VMIN and VMAX, the value of the corresponding limit is assigned.
The values VMIN and VMAX may be determined iteratively.
During an iteration, a value 64 of the y-axis 62 is selected in going through the axis from the bottom to the top. The smallest value VMIN and the 3o greatest value VMAX of the y-axis 61, corresponding to the intersections 64a, 64b of the histogram 60 with the selected value 64, are determined. The operation is stopped when the number of pixels to be eliminated for the selected value is greater than the determined fraction of the total number of pixels of the image.
The values taken by the component of the monochromatic image (the most discriminatory component) are distributed in a non-normalized range. In preparation for subsequent processing operations (such as segmentation, viewing, etc.), these values may be brought by linear transformation into a normalized interval, for example [0.1], or [0.255 ].
This linear transformation 65 may be applied, as illustrated in figure 6, after the pixels have been filtered in small numbers at the two limits of the histogram.
Thus, for an image encoded on 8 bits, the value VMIN is brought to a level zero, and the value VMAX is brought to a level 255.
Referring now to figure 7, an advantageous variant is described of the adjustment of the values used to process the image by giving greater or lesser importance to the contrast in a selected zone.
In this advantageous variant, the beginning of the processing is carried out in a similar way. The parameters of the Karhunen-Loeve ~5 transformation (or a linear approximation of this transformation) are determined according to the selected zone of the image only. This transformation is then applied to the entire image. The pixels of the selected zone occupy only a part 70 of the histogram 60.
Similarly, a linear transformation can be applied to bring the values of the pixels into a normalized interval. Advantageously, the parameters of the linear transformation are determined so as to bring the minimum value VMIN and the maximum value VMAX of the selected zone between two predetermined levels NMIN and NMAX. These levels NMIN and NMAX are strictly contained in the normalized interval. For example, for an image encoded on 8 bits, the normalized interval being [0.255 ], a value of 32 can be taken for NMIN and 224 for NMAX.
These parameters are used to apply a linear transformation to the entire image, the levels of the pixels being limited (for pixels outside the selected zone) in order to be included in the normalized interval.
3o This advantageous variant is used to obtain efficient discrimination in the regions similar to the selected zone, while at the same time keeping other information out of this selected zone.
This linear transformation 65 may be applied, as illustrated in figure 6, after the pixels have been filtered in small numbers at the two limits of the histogram.
Thus, for an image encoded on 8 bits, the value VMIN is brought to a level zero, and the value VMAX is brought to a level 255.
Referring now to figure 7, an advantageous variant is described of the adjustment of the values used to process the image by giving greater or lesser importance to the contrast in a selected zone.
In this advantageous variant, the beginning of the processing is carried out in a similar way. The parameters of the Karhunen-Loeve ~5 transformation (or a linear approximation of this transformation) are determined according to the selected zone of the image only. This transformation is then applied to the entire image. The pixels of the selected zone occupy only a part 70 of the histogram 60.
Similarly, a linear transformation can be applied to bring the values of the pixels into a normalized interval. Advantageously, the parameters of the linear transformation are determined so as to bring the minimum value VMIN and the maximum value VMAX of the selected zone between two predetermined levels NMIN and NMAX. These levels NMIN and NMAX are strictly contained in the normalized interval. For example, for an image encoded on 8 bits, the normalized interval being [0.255 ], a value of 32 can be taken for NMIN and 224 for NMAX.
These parameters are used to apply a linear transformation to the entire image, the levels of the pixels being limited (for pixels outside the selected zone) in order to be included in the normalized interval.
3o This advantageous variant is used to obtain efficient discrimination in the regions similar to the selected zone, while at the same time keeping other information out of this selected zone.
Claims (7)
1. A method of color image processing, comprising the following steps:
- the color image is converted into an intermediate image comprising components that depend solely on the H and S components of the original image in an HSV or HLS representation;
- the intermediate image comprising at least two components X and Y, determined by functions especially of the H component, these functions taking the same value when the H component is zero or equal to one;
- a new image comprising only one component is generated from the intermediate image, the component of this new image being a function of the components of the intermediate image.
- the color image is converted into an intermediate image comprising components that depend solely on the H and S components of the original image in an HSV or HLS representation;
- the intermediate image comprising at least two components X and Y, determined by functions especially of the H component, these functions taking the same value when the H component is zero or equal to one;
- a new image comprising only one component is generated from the intermediate image, the component of this new image being a function of the components of the intermediate image.
2. A method of color image processing according to the preceding claim, wherein the components X and Y are determined by functions not only of the H component, but also of the S component, these functions tending towards zero when the S component tends towards zero.
3. A method of color image processing according to the preceding claim, wherein the functions of the component S are monotonic and continuous functions.
4. A method of color image processing according to any of the preceding claims, wherein the new image is generated in keeping only the first component of the Karhunen-Loève transformation or a linear approximation of this transformation.
5. A method of color image processing according to any of the claims 1 to 3, wherein the new image is generated by projecting the components of the intermediate image in the plane in which the dynamic range or mean standard deviation is the greatest.
6. A method of color image processing according to any of the preceding claims, wherein a filtering is performed on the darkest and the lightest pixels, which represent a determined fraction of the total number of pixels of the image.
7. A method of color image processing according to any of the preceding claims, wherein the dynamic range of the new image is adjusted to the total available dynamic range.
Applications Claiming Priority (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
FR0211684A FR2844898B1 (en) | 2002-09-20 | 2002-09-20 | COLOR IMAGE PROCESSING METHOD FOR REMOVING SHADES AND REFLECTIONS. |
FR0211684 | 2002-09-20 | ||
PCT/EP2003/050403 WO2004027710A1 (en) | 2002-09-20 | 2003-09-12 | Method of color image processing to eliminate shadows and reflections. |
Publications (2)
Publication Number | Publication Date |
---|---|
CA2504567A1 true CA2504567A1 (en) | 2004-04-01 |
CA2504567C CA2504567C (en) | 2016-05-31 |
Family
ID=31970866
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CA2504567A Expired - Lifetime CA2504567C (en) | 2002-09-20 | 2003-09-12 | Method of color image processing to eliminate shadows and reflections |
Country Status (6)
Country | Link |
---|---|
US (1) | US7613353B2 (en) |
EP (1) | EP1543473A1 (en) |
AU (1) | AU2003271758A1 (en) |
CA (1) | CA2504567C (en) |
FR (1) | FR2844898B1 (en) |
WO (1) | WO2004027710A1 (en) |
Families Citing this family (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7751594B2 (en) | 2003-04-04 | 2010-07-06 | Lumidigm, Inc. | White-light spectral biometric sensors |
US8229185B2 (en) * | 2004-06-01 | 2012-07-24 | Lumidigm, Inc. | Hygienic biometric sensors |
US8787630B2 (en) * | 2004-08-11 | 2014-07-22 | Lumidigm, Inc. | Multispectral barcode imaging |
CN100334593C (en) * | 2006-03-07 | 2007-08-29 | 上海大学 | Shadow zone cutting method of two-dimensional color scene |
US7995808B2 (en) | 2006-07-19 | 2011-08-09 | Lumidigm, Inc. | Contactless multispectral biometric capture |
US8175346B2 (en) * | 2006-07-19 | 2012-05-08 | Lumidigm, Inc. | Whole-hand multispectral biometric imaging |
US8355545B2 (en) * | 2007-04-10 | 2013-01-15 | Lumidigm, Inc. | Biometric detection using spatial, temporal, and/or spectral techniques |
US8285010B2 (en) * | 2007-03-21 | 2012-10-09 | Lumidigm, Inc. | Biometrics based on locally consistent features |
TWI387354B (en) * | 2008-06-18 | 2013-02-21 | Novatek Microelectronics Corp | Method for adjusting color saturation and image displaying system using the method |
US20100246902A1 (en) * | 2009-02-26 | 2010-09-30 | Lumidigm, Inc. | Method and apparatus to combine biometric sensing and other functionality |
WO2011028620A1 (en) * | 2009-08-26 | 2011-03-10 | Lumidigm, Inc. | Multiplexed biometric imaging and dual-imager biometric sensor |
US8570149B2 (en) | 2010-03-16 | 2013-10-29 | Lumidigm, Inc. | Biometric imaging using an optical adaptive interface |
EP2821967A1 (en) * | 2013-07-03 | 2015-01-07 | Kapsch TrafficCom AB | Shadow detection in a multiple colour channel image |
CN109919963B (en) * | 2019-03-14 | 2023-03-24 | 吉林大学 | Vehicle paint defect position detection method |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH08186727A (en) * | 1994-12-28 | 1996-07-16 | Canon Inc | Device and method for processing picture |
JP3607059B2 (en) * | 1997-09-19 | 2005-01-05 | コニカミノルタビジネステクノロジーズ株式会社 | Color feature extraction apparatus, color feature extraction method, and program storage medium |
US7194128B1 (en) * | 2000-07-26 | 2007-03-20 | Lockheed Martin Corporation | Data compression using principal components transformation |
JP4398153B2 (en) * | 2001-03-16 | 2010-01-13 | ヴィジョン・ロボティクス・コーポレーション | Apparatus and method for increasing the effective dynamic range of an image sensor |
JP3943973B2 (en) * | 2002-03-20 | 2007-07-11 | キヤノン株式会社 | Image processing apparatus and method |
-
2002
- 2002-09-20 FR FR0211684A patent/FR2844898B1/en not_active Expired - Lifetime
-
2003
- 2003-09-12 WO PCT/EP2003/050403 patent/WO2004027710A1/en not_active Application Discontinuation
- 2003-09-12 AU AU2003271758A patent/AU2003271758A1/en not_active Abandoned
- 2003-09-12 US US10/528,447 patent/US7613353B2/en active Active
- 2003-09-12 EP EP03753590A patent/EP1543473A1/en not_active Ceased
- 2003-09-12 CA CA2504567A patent/CA2504567C/en not_active Expired - Lifetime
Also Published As
Publication number | Publication date |
---|---|
US7613353B2 (en) | 2009-11-03 |
US20060045330A1 (en) | 2006-03-02 |
CA2504567C (en) | 2016-05-31 |
EP1543473A1 (en) | 2005-06-22 |
FR2844898A1 (en) | 2004-03-26 |
AU2003271758A1 (en) | 2004-04-08 |
WO2004027710A1 (en) | 2004-04-01 |
FR2844898B1 (en) | 2005-03-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Kahu et al. | Review and evaluation of color spaces for image/video compression | |
Plataniotis et al. | Color image processing and applications | |
US7613353B2 (en) | Method of color image processing to eliminate shadows and reflections | |
US7593570B2 (en) | Image processing method and apparatus and storage medium | |
EP0723364B1 (en) | Real-time image enhancement techniques | |
US6519362B1 (en) | Method of extracting text present in a color image | |
US7468812B2 (en) | Image processing apparatus and its method for color correction | |
Yang et al. | Reduction of color space dimensionality by moment-preserving thresholding and its application for edge detection in color images | |
EP0878090A1 (en) | Appearance-based technique for rendering colors on an output device | |
JP3205836B2 (en) | Color quantization method using color coordinate space structure, color coordinates and color spraying | |
US8526719B2 (en) | Method of converting color image into grayscale image and recording medium storing program for performing the same | |
JP2003125423A (en) | Method for digital compression of color image | |
US6882449B2 (en) | Space varying gamut mapping | |
WO2001043076A1 (en) | Color conversion matrix based on minimal surface theory | |
KR100307822B1 (en) | Method for quantizing color using color coordinates | |
WO1997028640A1 (en) | Appearance-based technique for rendering colors on an output device | |
Hague et al. | Histogram equalization of the saturation component for true-color images using the CY color space | |
CN115168629B (en) | Image data compression storage method based on block chain | |
Plataniotis et al. | Color spaces | |
JPH01168167A (en) | Noise rejection method in multi-color picture | |
Enderlein | Classification of optical surface properties using color transformation to separate highly correlated bands | |
Haslam et al. | Color Restoration Survey and an Overdetermined System for Color Retrieval from Faded Images | |
JPS63108865A (en) | Color image processor and its manufacture | |
Yang | Effects of coordinate systems on color image processing | |
Sreelekshmi | Survey on His Image Fusion Methods |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
EEER | Examination request | ||
MKEX | Expiry |
Effective date: 20230912 |