Perceptual thresholding for gradient-based local edge detection

• US 6,094,508 A
• Filed: 12/08/1997
• Issued: 07/25/2000
• Est. Priority Date: 12/08/1997
• Status: Expired due to Term

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Tsai, Ping-Sing, Acharya, Tinku
• Primary Examiner(s): BELLA, MATTHEW C
• Application Number:
  US08/986,461
• Time in Patent Office: 960 Days
• Field of Search: 382/199, 382/201, 382/205, 382/256, 382/260, 382/261, 382/262, 382/263, 382/264, 382/265, 382/266, 382/270, 382/272, 382/274, 382/203
• US Class Current: 382/199
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 2207/20192 Edge enhancement; Edge preservation
• CPC Class Codes: G06T 7/12 Edge-based segmentation

Abstract

What is disclosed is a method for dynamically and automatically determining the threshold value for edge detection based on localized intensity information. Each pixel can be assigned a threshold to which its gradient or other intensity change information can be compared so as to detect whether or not the pixel belongs to an edge feature of the image or not. Further, the thresholding can be based upon the response of the human vision system so that visual appearance of the image to users is considered when analyzing localized intensity information. The localized thresholding technique can be applied to any edge detection scheme.

Claims

1. A method comprising:

defining a localization region of an image, said localization region including a selected portion of said image; and

determining automatically, without user intervention, a threshold applied for edge detection within said localization region based on intensity information of pixels within said localization region, said threshold determined by a power function operator reflective of perceptual characteristics as it relates to intensity of a human vision system, said power function operator is a measure of relative localization intensity raised to a power factor;

wherein said relative localization intensity is a quotient of a difference between a mean and a minimum of all intensity values within said localization region and a difference between a maximum and said minimum of all intensity values within said localization region.

2. A method according to claim 1 wherein said power factor corresponds to a perceptual ability of the human vision system to changes in intensity.

3. A method according to claim 2 wherein said power factor is 0.45.

4. A method according to claim 2 wherein said first mask is a Prewitt mask having coefficients of -1, -2, +1 and +2 variously in said second direction.

5. A method according to claim 1, further comprising:

detecting edge features of said image by comparing said threshold with a normalized gradient for a pixel P about which said localization region is defined.

6. A method according to claim 5 wherein said pixel is classified as a pixel belonging to an edge feature of said image if said normalized gradient of said pixel exceeds said threshold.

7. A method according to claim 5 wherein said normalized gradient is determined by performing the:

determining said gradient of every pixel within said localization region;

determining said maximum gradient value within said localization region; and

normalizing said gradient value of pixel P by dividing said gradient of pixel P by said maximum gradient value within said localization region.

8. A method according to claim 7 wherein a gradient of a given pixel is determined by:

applying a first mask to a neighborhood around each pixel whose gradient is being determined to attain a first direction differential quantity;

applying a second mask to said neighborhood around each pixel whose gradient is being determined to attain a second direction differential quantity, said second direction, wherein said second direction is perpendicular to said first direction; and

determining said gradient to be said magnitude of a vector defined by vector addition of said first direction differential quantity and said second direction differential quantity.

9. A method according to claim 10 wherein said magnitude is obtained by:

squaring said first direction differential;

squaring said second direction differential;

summing said squared first and second direction differentials; and

extracting said square root of said sum, said magnitude set to equal said square root.

10. A method according to claim 8 wherein said first mask is a Sobel mask having coefficients of -1 and +1 in variously said first direction.

11. A method according to claim 8 wherein said first mask is a Sobel mask having coefficients of -1 and +1 variously in said second direction.

12. A method according to claim 8 wherein said first mask is a Prewitt mask having coefficients of -1, -2, +1 and +2 variously in said first direction.

13. An article comprising a computer readable medium including instructions stored thereon that when executed by a processor, cause:

defining a localization region of an image, said localization region including a selected portion of said image; and

determining automatically, without user intervention, a threshold applied for edge detection within said localization region based on intensity information of pixels within said localization region, said threshold determined by a power function operator reflective of perceptual characteristics as it relates to intensity of a human vision system, said power function operator is a measure of relative localization intensity raised to a power factor;

wherein said relative localization intensity is a quotient of a difference between a mean and a minimum of all intensity values within said localization region and a difference between a maximum and said minimum of all intensity values within said localization region.

14. The article according to claim 13 wherein said power factor corresponds to the perceptual ability of the human vision system to changes in intensity.

15. The article according to claim 14 wherein said power factor is 0.45.

16. The article according to claim 13 where the computer readable medium further includes instructions that when executed by the processor cause:

detecting edge features of said image by comparing said threshold with a normalized gradient for a pixel P about which said localization region is defined.

17. The article according to claim 16 wherein said pixel is classified as a pixel belonging to an edge feature of said image if said normalized gradient of said pixel exceeds said threshold.

18. The article according to claim 16 wherein said normalized gradient is determined by

determining said gradient of every pixel within said localization region;

determining said maximum gradient value within said localization region; and

normalizing said gradient value of pixel P by dividing said gradient of pixel P by said maximum gradient value within said localization region.

19. The article according to claim 18, wherein a gradient of a given pixel is determined by:

applying a first mask to a neighborhood around each pixel whose gradient is being determined to attain a first direction differential quantity;

applying a second mask to said neighborhood around each pixel whose gradient is being determined to attain a second direction differential quantity, said second direction, wherein said second direction is perpendicular to said first direction; and

determining said gradient to be said magnitude of a vector defined by vector addition of said first direction differential quantity and said second direction differential quantity.

20. The article according to claim 19, wherein said magnitude is obtained by squaring said first direction differential;

squaring said second direction differential;

summing said squared first and second direction differentials; and

extracting said square root of said sum, said magnitude set to equal said square root.

21. The article according to claim 19, wherein said first mask is a Sobel mask having coefficients of -1 and +1 in variously said first direction.

22. The article according to claim 19, wherein said first mask is a Sobel mask having coefficients of -1 and +1 variously in said second direction.