Image processing methods and apparatus for detecting human eyes, human face, and other objects in an image

• US 20020081032A1
• Filed: 09/14/2001
• Published: 06/27/2002
• Est. Priority Date: 09/15/2000
• Status: Active Grant

• Current Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Original Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Inventors: Ji, Xin, Chen, Xinwu
• Granted Patent: US 6,965,684 B2
• Time in Patent Office: Days
• US Class Current: 382/199
• CPC Class Codes: G06T 2207/20132 Image cropping
• CPC Class Codes: G06T 2207/30201 Face
• CPC Class Codes: G06T 7/12 Edge-based segmentation
• CPC Class Codes: G06V 40/161 Detection; Localisation; Normalisation

Abstract

Disclosed is a method, apparatus, and system for detecting a human face in an image. The method includes, for a subset of pixels in the image, deriving a first variable from the gray-level distribution of the image and deriving a second variable from a preset reference distribution that is characteristic of the object. The method further includes evaluating the correspondence between the first variable and the second variable over the subset of pixels and determining if the image contains the object based on the result of this evaluation.

Claims

1. A method for detecting an object in an image having a gray-level distribution, characterized in that said method comprises the steps of:

a)for a subset of pixels in said image, deriving a first variable from said gray-level distribution of said image,

b)for said subset of pixels, deriving a second variable from a preset reference distribution, said reference distribution being characteristic of said object;

.

c)evaluating the correspondence between said first variable and said second variable over the subset of pixels; and

d)determining if said image contains said object based on the result of the evaluation step.

2. The method of claim 1, characterized by further comprising the step of:

selecting an area in said image,

wherein pixels of said subset are in said area, and

wherein over pixels in said area the first variable as derived from the gray-level distribution corresponding to the object to be detected has good correspondence to the second variable.

3. The method of claim 2, characterized in that it further comprises the step of:

determining a sub-image in said image; and

selecting said area on the basis of the location of said sub-image.

4. The method of any one of claim 1-3, wherein said first variable represents the direction of thee gradient of the gray-level distribution of said image.

5. The method of any one of claim 4, wherein said second variable represents the direction of the gradient of said reference distribution.

6. The method of claim 5, wherein said gradient is a discrete gradient.

7. The method of claim 5, characterized in that said evaluation step, includes a statistical process over said subset of pixels.

8. The method of claim 5, characterized in that said evaluation step includes a statistical processing over said subset of pixels and said statistical process comprises a weighting process, in which the weight to each pixel in said subset is determined on the basis of the gradient of the gray level distribution at the pixel and pixels with larger magnitude of the gradient of the gray level distribution generally have larger weights.

9. The method of claim 8, characterized in that said evaluating step includes both the statistical process using the weighting process and another statistical process without the weighting process and determines whether the image contains the object in accordance with the results of both of the statistical processes.

10. The method of claim 3, characterized in that said area is an annular area, which centers around the center of said sub-image.

11. The method of claim 2, wherein said subset of pixel includes all the pixels in said area.

12. The method of claim 3, characterized in that said sub-image is determined by detecting a predetermined feature in said image.

13. The image of claim 12, characterized in that said predetermined feature includes a pair of dark areas.

14. The method of claim 13, characterized in that said sub-image is determined based on the positions of and interval between the centers of said pair of dark areas.

15. The method of claim 13, characterized in that said reference distribution is expressed by

z(x,y)=−

(x²/a²+y²/b²)+h

wherein h is a constant, a/b equals to the ratio of the width of said sub-image to the height of said sub-image, and the origin is at the center of said sub-image.

16. The method of claim 5 or 15, wherein said evaluating step includes:

for all the pixels in said subset, calculating the average value of the angle between the gradient of the gray-level distribution of said image and the gradient of said reference distribution;

comparing said average value with a predetermined value; and

determining that the sub-image contains said object if said average value equals to or is smaller than said predetermined value.

17. The method of claim 5 or 15, wherein said evaluating step includes:

for all the pixels in said subset, calculating a weighted average value of the angle between the gradient of the gray-level distribution of said image and the gradient of said reference distribution, wherein a pixel having a greater magnitude of gradient of gray-level distribution generally has a larger weight;

comparing said weighted average value with a predetermined value; and

determining that the sub-image contains said object if said weighted average value equals to or is smaller than said predetermined value.

18. The method of claim 5 or 15, wherein said evaluating step includes:

for all the pixels in said subset, calculating the average value of the angle between the gradient of the gray-level distribution of said image and the gradient of said reference distribution;

for all the pixels in said subset, calculating a weighted average value of the angle between the gradient of the gray-level distribution of said image and the gradient of said reference distribution, wherein a pixel having a greater magnitude of gradient of gray-level distribution has a larger weight;

comparing said average value with a first predetermined value;

comparing said weighted average value with a second predetermined value; and

determining that the sub-image contains said object if said average value equals to or is smaller than said first predetermined value and said weighted average value equals to or is smaller than said second predetermined value.

53. The method according to claim 14, wherein the step for determining a sub-image in said image includes:

segmenting each column of the image into a plurality of intervals, labeling each of the intervals as valley region , intermediate region or peak region;

merging the valley region of the each column and the valley region of its adjacent column, thereby generating a candidate dark area; and

determining a dark area.

54. The method according to claim 53, wherein the merging step comprising the steps of

setting each valley region in the first column of the image as a seed region;

determining whether the valley region of the next column of the image can be merged into the seed region,

merging the valley region that can be merged into the seed region;

setting the valley region that cannot be merged into the seed region as a seed region; and

determining the seed region which has no further merged valley region as a dark area.

116. A storage device storing instructions for causing a programmable processing apparatus to become operable to perform a method as set out in at least one of claims 1 to 19, and 21 to 54.

117. A signal carrying instructions for causing a programmable processing apparatus to become operable to perform a method am sat out in at least one of claims 1 to 19, and 21 to 54.

Disclosed is a method, apparatus, and system for detecting human face in an image. The method comprises:

for a subset of pixels in said image, deriving a first variable from said gray-level distribution of said image;

for said subset of pixels, deriving a second variable from a preset reference distribution, said reference distribution being characteristic of said object, evaluating the correspondence between said first variable and said second variable over the subset of pixels;

determining if said image contains said object based on tie result of the evaluation step.

19. A method for detecting an object in an image having a gray-level distribution, characterized in that said method comprises the steps of:

a) determining a sub-image in said image;

b) selecting a subset of pixels in said image based on said sub-image;

c) for pixels of said subset, deriving a first variable from said gray-level distribution of said image,

d) for said subset of pixels, deriving a second variable from a preset reference distribution, said reference distribution being characteristic of said object;

e) evaluating the correspondence between said first variable and said second variable over the subset of pixels; and

f) determining if said image contains said object based on the result of the evaluation step.

20. A storage medium with a program code stored therein for detecting an object in an image with a gray-level distribution, characterized in that said program code comprises:

codes for determining a sub-image in said image;

codes for deriving, for a subset of pixels, a first variable from said gray-level distribution of said image,

codes for deriving, for said subset of pixels, a second variable from a preset reference distribution, said reference distribution being characteristic of said object;

codes for evaluating the correspondence between said first variable and said second variable over the subset of pixels; and

codes for determining if said image contains said object based on the result of the evaluation step.

21. An image processing method for determining a feature portion in an image, comprising:

a reading step of reading the image and a rectangle region to be determined in the image;

a setting step of setting a annular region surrounding said rectangle region;

a step of calculating the gradient of gray level of each pixel in said annular region,

a step of determining a reference distribution for the annular region and determining a annular region reference gradient of each pixel in the annular region; and

a determination step of determining the feature portion contained in said rectangle region on the basis of the gradient of the gray level of each pixel and the reference gradient of each pixel in said annular region.

22. The image processing method according to claim 21, wherein the image containing said rectangle region to be determined is a feature portion.

23. The image processing method according to claim 21, wherein said annular region is formed by a first rectangle region and a second rectangle region.

24. The image processing method according to claim 23, wherein said first rectangle region is located inside said rectangle region to be determined.

25. The image processing method according to claim 23, wherein said second rectangle region is located outside said rectangle region to be determined.

26. The image processing method according to claim 23, wherein the center of said first rectangle region lies in substantially the same location as that of said rectangle region to be determined, and the width and the length of said first rectangle region respectively equals m times the width and the length of said rectangle region to be determined, where 0<

m<

1.

27. The image processing method according to claim 23, wherein the center of said second rectangle region lies in the same location as that of said rectangle region to be determined, and the width and the length of said second rectangle region respectively equals n times the width and the length of said rectangle region to be determined, where 1<

n<

2.

28. The image processing method according to claim 21, wherein said step of calculating the gradient of gray level comprising the step of calculating the gradient of the gray level of a pixel on the basis of the gray level of the pixels surrounding the pixel.

29. The image processing method according to claim 28, wherein the number of pixels surrounding said pixel equals k², where k is an integer, and 2<

k<

15.

30. The image processing method according to claim 21, wherein reference distribution is determined in accordance to the rectangle region.

31. The image processing method according to claim 21 or 30, further comprising the step of determining the angle the gradient of the gray level and the reference gradient of each pixel in said annular region.

32. The image processing method according to claim 21 or 30, further comprising the step of determining the average of the angle between the gradient of the gray level of each pixel and the reference gradient of each pixel in said annular region.

33. The image processing method according to claim 21 or 32, further comprising the step of determining the weight of the gradient of the gray level of each pixel in said annular region.

34. The image processing method according to claim 32, further comprising the step of determining whether the average of the angle between the gradient of the gray level of each pixel and the reference gradient of each pixel in said annular region is less than a first threshold, where the first threshold is within the range of 0.01 to 1.5, preferably of 0.61.

35. The image processing method according to claim 34, further comprising the step of determining whether the weighted average of the angle between the gradient of the gray level of each pixel and the reference gradient of each pixel in said annular region with the weight of the gradient of the gray level of each pixel in said annular region is less than a second threshold, where the second threshold is within the range of 0.01 to 1, preferably of 0.68.

36. The image processing method according to claim 33, further comprising the step of determining whether the weighted average of the angle between the gradient of the gray level of each pixel and the reference gradient with the weight of the gradient of the gray level of each pixel in said annular region is less than a third threshold, where the third threshold is within the range of 0.1 to 1, preferably of 0.68.

38. The method according to claim 37, wherein the segmenting step includes the step of labeling each interval as one of valley region, intermediate region and peak region based on the gray level of each the interval in a column.

39. The method according to claim 37, wherein the segmenting step includes the step of labeling each interval as one of valley region, intermediate region and peak region based on the luminance value of each interval in a column.

40. The method according to claim 37, wherein the segmenting step includes the step of labeling each interval in a column as one of valley region, intermediate region and peak region based on the respective gray level ratios of the interval to its two adjacent intervals.

41. The method according to claim 37, wherein the segmenting step includes the step of labeling each interval in a column as one of valley region, intermediate region and peak region based on the respective luminance value ratios of the interval to its two adjacent interval.

44. The method according to claim 37, wherein the merging step comprising the steps of

setting each valley region in the first column of the image as a seed region respectively;

determining whether the valley region of next column of the image can be merged into the seed region;

merging the valley region that can be merged into the seed region;

setting the non-merged valley region as a seed region; and

determining the seed region which has no further merged valley region as an eye candidate region.

45. The method according to claim 44, further comprising the step of comparing the size of overlap of the valley region and the seed region'"'"'s predicted valley region, the valley region and the seed region'"'"'s predicted valley region,

46. The method according to claim 44, further comprising the step of comparing the gray level of the valley region and that of the seed region.

47. The method according to claim 44, further comprising the step of comparing the gray levels of the valley region, the seed region, the background gray levels of the valley region and the seed region.

48. The method according to claim 37, wherein the determining step comprising the step of determining whether the gray level of the eye candidate region is less than the first threshold, where the first threshold is within the range of 100 to 200, preferably of 160.

49. The method according to claim 48, wherein the determining step further comprising the step of determining whether the background gray level of the eye candidate region is bigger than the second threshold, where the second threshold is within the range of 20 to 80, preferably of 30.

50. The method according to claim 49, wherein the determining step further comprising the step of determining whether the difference of background gray level of the eye candidate region and its gray level is bigger than a third threshold, where the third threshold is within the range of 5 to 120, preferably of 20.

51. The method according to claim 50, wherein the determining step further comprising the step of determining whether the ratio of the width to the height of an eye candidate region is bigger than a fourth threshold, where the fourth threshold is within the range of 1 to 5, preferably of 3.33.

52. The method according to claim 51, wherein the determining step further comprising the step of determining whether the ratio of the size of an eye candidate region to that of its circum-rectangle is bigger than the fifth threshold, where the fifth threshold is within the range of 0.2 cm to 1 preferably of 0.4.

37. A human eye detecting method for detecting human eye in an image, comprising:

a reading step of reading the gray level of each pixel in the each column in the image;

a segmenting step of segmenting each column into a plurality of intervals, and labeling each of the intervals as valley region intermediate region or peak region;

a merging step of merging the valley region of the each column and the valley region of its adjacent column, and generating an eye candidate region; and

a determining step of determining the human eye from the eye candidate regions.

42. The method according to any one of claims 37, further comprising the step of comparing the gray level of a segment point with that of its two adjacent intervals, and adding the segment point into the adjacent interval which has a gray level close to that of the segment point.

43. The method according to any one of claims 37, further comprising the step of comparing the gray level of a intermediate region with that of its adjacent valley region and peak valley, and adding the intermediate region into the valley region on the basis of the gray levels of the adjacent valley region, peak region and the intermediate region.

55. Apparatus for detecting an object in an image having a gray-level distribution, comprising:

a) means for deriving a first variable from said gray-level distribution of said image for a subset of pixels in said image,

b) means for deriving a second variable from a preset reference distribution for said subset of pixels, amid reference distribution being characteristic of said object;

c) correspondence evaluating means for evaluating the correspondence between said first variable and said second variable over the subset of pixels; and

d) means for determining if said inage contains said object based on the processing by the correspondence evaluating means.

56. The apparatus of claim 55, further comprising means for selecting an area in said image, and wherein:

pixels of said subset are in said area, and

over pixels in said area, the first variable as derived from the gray-level distribution corresponding to the object to be detected has good correspondence to the second variable.

57. The apparatus of claim 56. further comprising:

means for determining a sub-image in said image; and

means for selecting said area on the basis of the location of said sub-image.

58. The apparatus of any one of claim 55-57, wherein said first variable represents the direction of the gradient of the gray-level distribution of said image.

59. The apparatus of claim 58, wherein said second variable represents the direction of the gradient of said reference distribution.

60. The apparatus of claim 59, wherein said gradient is a discrete gradients.

61. The apparatus of claim 59, wherein the correspondence evaluating means is arranged to perform a statistical process over said subset of pixels.

62. The apparatus of claim 59, wherein the correspondence evaluating means is arranged to perform statistical processing over said subset of pixels and said statistical process comprises a weighting process, in which a weight to each pixel in said subset is determined on the basis of the gradient of the gray level distribution at the pixel and pixels with larger magnitude of the gradient of the gray level distribution generally have larger weight.

63. The apparatus of claim 62, wherein the correspondence evaluating means is arranged to perform processing that includes both the statistical process using the weighting process and another statistical process without the weighting process, and is arranged to determine whether the image contains the object in accordance with the results of both of the statistical processes.

64. The apparatus of claim 57, wherein said area is an annular area, which centers around the center of said sub-image.

65. The apparatus of claim 56, wherein said subset of pixel includes all the pixels in said area.

66. The apparatus of claim 57, wherein the means for determining a sub-image is arranged to determine said sub-image by detecting a predetermined feature in said image.

68. The apparatus of claim 67, wherein the means for detecting a sub-image is arranged to determine said sub-image based on the positions of and interval between the centers of said pair of dark areas.

69. The apparatus of claim 67, wherein said reference distribution is expressed by

z(x,y)=−

(x²/a²+y²/b²)+h

wherein h is a constant, a/b is the ratio of the width of said sub-image to the height of said sub-image, and the origin is at the center of said sub-image.

70. The apparatus of claim 59 or 69, wherein said correspondence evaluating means comprises:

means for calculating, for all the pixels in said subset, the average value of the angle between the gradient of the gray-level distribution of said image and the gradient of said reference distribution; and

means for comparing said average value with a predetermined value;

and wherein the means for determining if the image contains the object is arranged to determine that the sub-image contains said object if said average value is equal to or is smaller than said predetermined value.

71. The apparatus of claim 59 or 69, wherein said correspondence evaluating means comprises:

means for calculating, for all the pixels in said subset, a weighted average value of the angle between the gradient of the gray-level distribution of said image and the gradient of said reference distribution, wherein a pixel having a greater magnitude of gradient of gray-level distribution generally has a larger weight; and

means for comparing said weighted average value with a predetermined value;

and wherein the means for determining if the image contains the object is arranged to determine that the sub-image contains said object if said weighted average value is equal to or is smaller than said predetermined value.

72. The apparatus of claim 59 or 69, wherein said correspondence evaluating means comprises;

means for calculating, for all the pixels in said subset, the average value of the angle between the gradient of the gray-level distribution of said image and the gradient of said reference distribution;

means for calculating, for all the pixels in said subsets a weighted average value of the angle between the gradient of the gray-level distribution of said image and the gradient of said reference distribution, wherein a pixel having a greater magnitude of gradient of gray-level distribution has a larger weight;

means for comparing said average value with a first predetermined value; and

means for comparing said weighted average value with a second predetermined value;

and wherein the means for determining iv the image contains the object is arranged to determine that the sub-image contains said object if said average value is equal to or is smaller than said fist predetermined value and said weighted average value is equal to or is smaller than said second predetermined value.

114. The apparatus according to claim 68, wherein the means for determining a sub-image in said image includes:

means for segmenting each column of the image into a plurality of intervals,

means for labeling each of the intervals as valley region, intermediate region or peak region;

means for merging the valley region of each column and the valley region of its adjacent column, thereby generating a candidate dark area; and

means for determining a dark area.

67. The apparatus of clam 66, wherein said predetermined feature includes a pair of dark areas.

73. Apparatus for detecting an object in an image having a gray-level distribution, comprising:

a) means for determining a sub-image in said image;

b) means for selecting a subset of pixels in said image based on said sub-image;

c) means for deriving, for pixels of said subset, a first variable from said gray-level distribution of said image;

d) means for deriving, for said subset of pixels, a second variable from a preset reference distribution, said reference distribution being characteristic of said object;

e) evaluating means for evaluating the correspondence between said first variable and said second variable over the subset of pixels; and

f) means for determining if said image contains said object based an the result of the processing by the evaluating means.

74. An image processing;

apparatus for determining a feature portion in a image comprising;

reading means for reading the image ad a rectangle region to be determined in the image;

setting means for setting an annular region surrounding said rectangle region;

gradient calculating means far calculating the gradient of gray level of each pixel in said annular region;

reference distribution determining means for determining a reference distribution for the annular region and determining an annular region reference gradient of each pixel in the annular region; and

feature portion determination means for determining the feature portion contained in said rectangle region on the basis of the gradient of the gray level of each pixel and tho reference gradient of each pixel in said annular region.

75. The apparatus according to claim 74, wherein the image containing said rectangle region to be determined is a feature portion.

76. The apparatus according to claim 74, wherein the setting means is arranged to form said annular region by a first rectangle region and a second rectangle region.

77. The apparatus according to claim 76, wherein the setting means is arranged to locate said first rectangle region inside said rectangle region to be determined.

78. The apparatus according to claim 76, wherein the setting means is arranged to locate said second rectangle region outside said rectangle region to be determined.

79. The apparatus according to claim 76, wherein the setting means is arranged to locate the center of said first rectangle region in substantially the same location as that of said rectangle region to be determined, and to set the first rectangle region so that the width and the length of said first rectangle region respectively equals m times the width and the length of said rectangle region to be determined, where 0<

m<

1.

80. The apparatus according to claim 76, wherein the setting means is arranged to locate the center of said second rectangle region in the same location as that of said rectangle region to be determined, and to set the second rectangle region so that the width and the length of said second rectangle region respectively equals n times the width and the length of said rectangle region to be determined, where 1<

n<

2.

81. The apparatus according to claim 74, wherein said gradient calculating means comprises means for calculating the gradient of the gray level of a pixel on the basis of the gray level of the pixels surrounding the pixel.

82. The apparatus according to claim 81, wherein the gradient calculating means is arranged to select the number of pixels surrounding said pixel to be used in the gradient calculation to be k², where k is an integer, and 2<

k<

15.

83. The apparatus according to claim 74, wherein the reference distribution determining means is arranged to determine said reference distribution in accordance with the rectangle region.

84. The apparatus according to claim 74 or 83, further comprising means for determining the angle of the gradient of the gray level and the reference gradient of each pixel in said annular region.

85. The apparatus according to claim 74 or 83, further comprising means for determining the average of the angle between the gradient of the gray level of each peel and the reference gradient of each pixel in said annular region.

86. The apparatus according to claim 74 or 85, further comprising means for determining the weight of the gradient of the gray level of each pixel in said annular region.

87. The apparatus according to claim 85, further comprising means for determining whether the average of the angle between the gradient of the gray level of each pixel and the reference gradient of each pixel in said annular region is less than a first threshold, where the first threshold is within the range of 0.01 to 1.5.

88. The apparatus according to claim 87, wherein the first threshold is 0.61.

89. The apparatus according to claim 87 or claim 88, further comprising means for determining whether the weighted average of the angle between the gradient of the gray level of each pixel and the reference gradient of each pixel in said annular region with the weight of the gradient of the gray level of each pixel in said annular region is less than a second threshold, where the second threshold is within the range of 0.01 to 1.

90. The apparatus according to claim 89, wherein the second threshold is 0.68.

91. The apparatus according to claim 86, further comprising means for determining whether the weighted average of the angle between the gradient of the gray level of each pixel and the reference gradient with the weight of the gradient of the gray level of each pixel in said annular region is less than a third threshold, where the third threshold is within the range of 0.1 to 1.

92. The apparatus according to claim 91 wherein the third threshold is 0.68.

93. Apparatus for detecting a human eye in an image, comprising:

reading means for reading the gray level of each pixel in the each column in the image;

segmenting means for segmenting each column into a plurality of intervals, and labeling each of the intervals as valley region, intermediate region or peak region;

merging means for merging the valley region of the each column and the valley region of its adjacent column, and generating an eye candidate region; and

determining means for determining if a human eye exists in the eye candidate regions.

94. The apparatus according to claim 93, wherein the segmenting means includes means far labeling each interval as one of valley region, intermediate region and peak region based on the gray level of each interval in a column.

95. The apparatus according to claim 93, wherein the segmenting means includes means fox labeling each interval as one of valley region, intermediate region and peak region based on the luminance value of each interval in a column.

96. The apparatus according to claim 93, wherein the segmenting means includes means or labeling each interval in a column as one of valley region, intermediate region and peak region based on the respective gray level ratios of the interval to its two adjacent intervals.

97. The apparatus according to claim 93, wherein the segmenting means includes means for labeling each interval in a column as one of valley region, intermediate region and peak region based on the respective luminance value ratios of the interval to its two adjacent intervals.

98. The apparatus according to claim 93, further comprising means for comparing the gray level of a segment point with that of its two adjacent intervals, and adding the segment point into the adjacent interval which has a gray level close to that of the segment point.

99. The apparatus according to claim 93, further comprising means for comparing the gray level of an intermediate region with that of its adjacent valley region and peak valley, and adding the intermediate region into the valley region on the basis of the gray levels of the adjacent valley region, beak region and the intermediate region.

100. The apparatus according to claim 93, wherein the merging means comprises:

means for setting each valley region in the first column of the image as a seed region respectively;

means for determining whether the valley region of a next column of the image can be merged into the seed region;

means for merging a valley region that can be merged into the seed region;

means for setting a non-merged valley region as a seed region; and

means for determining the seed region which has no further merged valley region as an eye candidate region.

101. The apparatus according to claim 100, further comprising means for comparing the size of overlap of the valley region and the seed region'"'"'s predicted valley region.

102. The apparatus according to claim 100, further comprising means for comparing the gray level of the valley region and that of the seed region.

103. The apparatus according to claim 100, further comprising means for comparing the gray levels of the valley region, the seed region, the background gray levels of the valley region and the seed region.

104. The apparatus according to claim 93, wherein the determining means comprises means for determining whether the gray level of the eye candidate region is less than a first threshold, where the first threshold is within the range of 100 to 200.

105. The apparatus according to claim 104, wherein the first threshold is 160.

106. The apparatus according to claim 104 or claim 105, wherein the determining means further comprises means for determining whether the background gray level of the eye candidate region is bigger than a second threshold, where the second threshold is within the range 20 to 80.

107. The apparatus according to claim 106, wherein the second threshold is 30.

108. The apparatus according to claim 106 or claim 107, wherein the determining means further comprises means for determining whether the difference of background gray level of the eye candidate region and its gray level is bigger than a third threshold, where the third threshold is within the range 5 to 120.

109. The apparatus according to claim 108, wherein the third threshold is 20.

110. The apparatus according to claim 108 or claim 109, wherein the detesting means further comprises means for determining whether the ratio of the width to the height of an eye candidate region is bigger than a fourth thresholds where the fourth threshold is within the range of 1 to 5.

111. The apparatus according to claim 110, wherein the fourth threshold is 3.33.

112. The apparatus according to claim 110 or claim 111, wherein the determining means further comprises means for determining whether the ratio of the size of an eye candidate region to that of its circum-rectangle is bigger than a fifth threshold, where the fifth threshold is within the range of 0.2 to 1.

113. The apparatus according to claim 112, wherein the fifth threshold is 0.4.

115. The apparatus according to claim 114, wherein the merging means comprises:

means for setting each valley region in the first column of the image as a seed region;

means for determining whether the valley region of the next column of the image can be merged into the seed region,

mean for setting the valley region that can be merged into the seed region; and

means for setting he valley region that cannot be merged into the seed region as a seed region;

and wherein the means for determining a dark area is arranged to determine a send region which has no further merged valley region as a dark area.