IMAGE PROCESSING BASED ON VISUAL ATTENTION AND REDUCED SEARCH BASED GENERATED REGIONS OF INTEREST

• US 20120070075A1
• Filed: 09/17/2010
• Published: 03/22/2012
• Est. Priority Date: 09/17/2010
• Status: Active Grant

• Current Assignee: Honeywell International Inc.
• Original Assignee: Honeywell International Inc.
• Inventors: Eswara, Lalitha M.
• Granted Patent: US 8,542,875 B2
• Time in Patent Office: Days
• US Class Current: 382/162
• CPC Class Codes: G06V 10/25 Determination of region of interest [ROI] or a volume of interest [VOI]

Abstract

A system for complexity reduction in images involving concepts of visual attention based most probable region detection for object presence and perspective-view based reduced scale-search approaches. Visual attention concept in context uses gradient and contrast of an image. A pixel meeting certain criteria for gradient or contrast values may be further processed for object presence. Limiting image processing to such regions may reduce the complexity of digitized images. Post processing the outcome using morphological operations like dilation and erosion appropriately may help retain some of the missed object pixels in the resultant image. Typically image blocks at different scales are searched for object presence. Reduced-scale search involves removing certain scales during search. As object size in image varies with its location scales within a given scale-range if searched, may lead to higher chances of object presence. This is implemented using relative heights and widths estimation using the perspective view concept.

Claims

1. A method for reducing complexity of processing digitized images for object detection using a reduced scale-search, comprising:

obtaining samples of object heights or widths at different locations in one or more images;

performing a non-linear interpolation of heights or widths for each pixel in an image (RScale);

considering an image block for processing based on a scale; and

determining whether a height or width of the image block being processed is greater than X percent of RScale of a pixel at an image block center and is less than Y percent of an RScale;

wherein;

RScale is a non-linear interpolation of heights or widths of an object for each pixel in the one or more images;

X is a number; and

Y IS A NUMBER.

2. The method of claim 1, wherein the method is performed only on pixels in the one or more images, that need to be processed.

3. The method of claim 1, wherein if an estimated height or width of the object at a center pixel in the image block being processed is greater than X percent of RScale of the center pixel of the block and is less than Y percent of the RScale, then the image block is processed for object presence.

4. The method of claim 1, wherein if the estimated height or width of the object at the center pixel in the image block being processed is not greater than X percent of RScale of the center pixel of the block or is not less than Y percent of the RScale, then the image block is not processed for object presence.

5. The method of claim 1, wherein RScale is re-computed whenever there is change in an object size of the object to be detected in an image, camera, or its placement, or its field of view or its view change.

6. An approach for reducing complexity of digitized images by using visual attention, comprising:

determining a gradient image of an input image;

determining a YUV image of the input image;

evaluating gradient pixels of the gradient image;

processing the gradient pixels having values meeting an adaptive threshold;

evaluating Y-channel pixels of the YUV image as contrast pixels;

processing contrast pixels having values meeting an adaptive threshold; and

merging processed gradient and contrast pixels into an output image.

7. The approach of claim 6, further comprising a post-processor to minimize missed pixels of an object in the output image where there are no processed pixels using morphological operations like erosion and dilation.

8. The approach of claim 6, wherein:

a contrast image comprises evaluated pixels of a Y-channel of the YUV image; and

the evaluated pixels of a Y-channel image are contrast pixels.

9. The approach of claim 8, further comprising:

determining a mean and standard deviation of values of the pixels of the Y-channel image; and

checking whether a value of a pixel of the Y-channel image is <

(M′

−

s*S′

) or >

(M′

+s*S′

); and

wherein;

M′

represents the mean;

S′

represents the standard deviation;

s represents a multiplication factor; and

the adaptive threshold is <

(M′

−

s*S′

) or >

(M′

+s*S′

).

10. The approach of claim 9, wherein:

if a value of a pixel of the Y-channel image is <

(M′

−

s*S′

) or >

(M′

+s*S′

), then the pixel is processed; and

if a value of a pixel of the Y-channel image is not <

(M′

−

s*S′

) or >

(M′

+s*S′

), then the pixel is not processed.

11. The approach of claim 6, further comprising:

determining a minimum and a maximum of values of the gradient pixels of the gradient image; and

checking whether a value of a gradient pixel is <

(S*m) or >

(S*M); and

wherein;

M represents the maximum;

m represents the minimum;

S represents a scale factor; and

the adaptive threshold is <

(S*m) or >

(S*M).

12. The approach of claim 11, wherein:

if a value of a gradient pixel is <

(S*m) or >

(S*M), then the pixel is processed; and

if a value of a gradient pixel is not <

(S*m) or >

(S*M), then the pixel is not processed.

13. A system for reducing complexity of digital images, comprising:

an image source module;

a gradient pixel selection module connected to the image source module;

a contrast pixel selection module connected to the image source module; and

a pixel merger module connected to the gradient pixel selection module and to the contrast pixel selection module.

14. The system of claim 13, wherein the gradient pixel selection module comprises:

a gradient detector connected to the image source module; and

a gradient pixel evaluator connected to the gradient pixel selection module and the pixel merger module.

15. The system of claim 13, wherein the contrast pixel selection module comprises:

a Color Component detector connected to the image source module; and

a contrast pixel evaluator connected to the color component detector and the pixel merger module.

16. The system of claim 13, wherein the pixel merger module comprises:

an image pixel screener or selector;

a post processor; and

a binary image output module connected to the post processor.

17. The system of claim 14, wherein the gradient pixel evaluator finds minimum and maximum values, and selects a scale factor for a gradient of an image.

18. The system of claim 17, wherein:

if a gradient pixel has a value less than a product of a scale factor and the minimum value or greater than a product of the scale factor and the maximum value, then the gradient pixel is processed as part of an output image; and

if the gradient pixel does not have a value less than a product of the scale factor and the minimum value or greater than a product of the scale factor and the maximum value, then the gradient pixel is not processed as part of an output image.

19. The system of claim 14, wherein the contrast pixel evaluator finds a mean, a standard deviation and selects a multiplication factor for a Y-channel image.

20. The system of claim 19, wherein:

if a Y-channel pixel has a value less than the mean minus a product of the multiplication factor and the standard deviation or greater than the mean plus the product of the multiplication factor and the standard deviation, then the pixel is processed; and

if the Y-channel pixel does not have a value less than the mean minus a product of the multiplication factor and the standard deviation or greater than the mean plus the product of the multiplication factor and the standard deviation, then the pixel is not processed as part of an output image.