Method for detecting color objects in digital images

• US 7,263,220 B2
• Filed: 02/28/2003
• Issued: 08/28/2007
• Est. Priority Date: 02/28/2003
• Status: Expired due to Fees

• Current Assignee: Monument Peak Ventures, LLC (Dominion Harbor Enterprises, LLC)
• Original Assignee: Eastman Kodak Company
• Inventors: Luo, Jiebo, Crandall, David J.
• Primary Examiner(s): Chen; Wenpeng
• Assistant Examiner(s): Ge; Yuzhen
• Application Number:
  US10/377,362
• Publication Number:
  US 20040170318A1
• Time in Patent Office: 1,642 Days
• Field of Search: 382/165
• US Class Current: 382/165
• CPC Class Codes: G06V 10/56 relating to colour
• CPC Class Codes: G06V 10/758 Involving statistics of pixels or of feature values, e.g. histogram matching

Abstract

A method for detecting an object in a digital image including the steps of performing color quantization on a model image including the object and on a search image that potentially includes the object, generating a plurality of search windows, computing spatial-color joint probability functions of each model and search image, where the color co-occurrence edge histogram is chosen to be the spatial-color joint probability function, assessing the similarity of each search image to the model, and designating search windows as containing the target object.

Claims

1. A method of finding a color object in a digital image, comprising the following steps:

(a) obtaining a search image that potentially includes the color object;

(b) capturing a single model image including a model of the color object;

(c) computing an edge Spatial Color Joint Probability Function (SCJPF) for the model image, wherein a model image edge SCJPF is a distribution of colors at color edges as well as spatial relationships between such colors in the model image, wherein a color edge is defined as a pixel whose color is different from its neighboring pixels according to a predetermined threshold;

(d) generating a plurality of search windows each comprising a portion of the search image;

(e) computing an edge SCJPF for each search window, wherein a search window SCJPF is a distribution of colors at color edges as well as spatial relationships between such colors in the search window;

(f) assessing a first degree of similarity between the model image edge SCJPF and each of the search window edge SCJPFs;

(g) computing a Spatial Color Joint Probability Function (SCJPF) for the model image, wherein a model image SCJPF is a distribution of colors as well as spatial relationships between such colors in the model image;

(h) computing a SCJPF for each search window, wherein a search window SCJPF is a distribution of colors as well as spatial relationships between such colors in the search window;

(i) assessing a second degree of similarity between the model image SCJPF and the search window SCJPF; and

(j) designating a search window associated with (1) search window edge SCJPFs having a first degree of similarity to the model image edge SCJPFs which exceeds a first prescribed search threshold and (2) a search window SCJPF having a second degree of similarity to a model image SCJPF which exceeds a second prescribed search threshold as potentially containing the object being sought.

2. The method as claimed in claim 1 further comprising the step of performing color quantization on both the model image and the search image, wherein the number of unique colors in the images is reduced to a set of pre-determined perceptually related color names.

3. The method as claimed in claim 2 wherein the color quantization step is further comprised of the steps of:

converting each pixel value in the model image and the search image to a suitable color space;

assigning each pixel a color name based on the location of its color value within the color space; and

assigning each pixel a quantized color value based on its assigned color name.

4. The method as claimed in claim 2 wherein the reduced set of pre-determined perceptually related color names includes human skin color.

5. The method as claimed in claim 2 wherein the step of performing color quantization further includes mapping the color of a pixel to more than one color name in the reduced set of pre-determined perceptually related color names.

6. The method as claimed in claim 1 wherein each search window is within a region of the image having a minimum predefined number of the colors present in the object model image.

7. The method as claimed in claim 1 wherein the step (1) of designating search windows employs a similarity assessment that is performed by mathematical correlation of the model image SCJPF and each of the search window SCJPFs, and the model image color histogram and each of the search window color histograms.

8. The method as claimed in claim 1 wherein the step (e) of computing a Spatial Color Joint Probability Function (SCJPF) for each search window further includes computing a first partial SCJPF for a part of the search window that does not overlap with a previously-visited search window, and adding the first partial SCJPF to a second partial SCJPF computed from the previously-visited search window.

9. The method as claimed in claim 1 further including a step of pre-screening the search image to identify regions that are more likely to contain the color object before generating a plurality of search windows in step (d), thereby eliminating regions that are less likely to contain the color object and that do not therefore warrant further processing.

10. The method of claim 9 wherein said pre-screening step includes the following:

identifying the important colors in object model image;

identifying pixels in search image which contain in their neighborhoods at least a predefined number of the important model image colors;

creating a binary mask of matching neighborhood centers;

performing connected component analysis; and

determining minimum enclosing rectangle of each connected component, the set of rectangles determined making up the set of search regions that may contain the target object.

11. The method of claim 9 wherein said pre-screening step includes the following:

computing the color co-occurrence histogram of object model image;

generating a probability map image by random examination of pixel pairs in search image and comparison to model image SCJPF;

performing low-pass filtering on probability map;

thresholding the probability map at a predefined level;

performing connected component analysis; and

determining the minimum enclosing rectangle of each connected component to generate the set of search regions.

12. A method of finding a color object in a digital, comprising the following steps:

obtaining a search image that potentially includes the color object;

capturing a single model image including a model of the color object;

generating a plurality of search windows each comprising a portion of the search image;

computing a probability function for the model image and for each of the search windows, said probability function being

an edge Spatial Color Joint Probability Function, wherein the model image probability function and search window probability function are each a distribution of colors at color edges as well as spatial relationships between such colors in the respective model image or search window, respectively, wherein a color edge is defined as a pixel whose color is different from its neighboring pixels according to a predetermined threshold;

assessing a degree of similarity between the model image probability function and each of the search window probability functions; and

designating search windows associated with the search window probability functions having a degree of similarity to the model image probability function which exceeds a prescribed search threshold as potentially containing the object being sought.