Diffusion based peer group processing method for image enhancement and segmentation

• US 6,408,028 B1
• Filed: 06/01/1998
• Issued: 06/18/2002
• Est. Priority Date: 06/02/1997
• Status: Expired due to Fees

• Current Assignee: Regents of the University of California (University of California)
• Original Assignee: Regents of the University of California (University of California)
• Inventors: Manjunath, Bangalore S., Kenney, Charles
• Primary Examiner(s): BRITTON, HOWARD W
• Application Number:
  US09/088,531
• Time in Patent Office: 1,478 Days
• Field of Search: 375/240.08, 382/243
• US Class Current: 375/240.08
• CPC Class Codes: G06T 2207/20192 Edge enhancement; Edge preservation
• CPC Class Codes: G06T 5/20 using local operators
• CPC Class Codes: G06T 5/70 Denoising; Smoothing
• CPC Class Codes: G06T 5/77 Retouching; Inpainting; Scratch removal

Abstract

An image processing method wherein a “peer group” is identified for each pixel in the image and the peer group is then used to make processing decisions at that pixel. A diffusion form of peer group processing is presented that averages over the peer group (PGA). Two parameters provide direct control over which objects are selectively enhanced: area (number of pixels in the object) and window diameter (window size needed to enclose the object). The direct correspondence between these parameters and the object characteristics is in sharp contrast to parameter selection for other image processing methods such as variational approaches in which it is unclear how to select good values for the smoothness and approximation weights. The scale space decomposition of an image determined by using the peer group number as the scale parameter leads to top hat procedures for recovering objects in a given size range.

Claims

1. A method for enhancing images having n or more pixels, comprising:

using a window diameter d and a peer group number n, selecting a peer group of size n for each pixel in an image from a window centered at the pixel, said peer group comprising the n pixels having values closest to the value of said center pixel; and

processing the pixel value based on a characteristic of the peer group.

2. A method as recited in claim 1, further comprising averaging the values of the peer group pixels to obtain a new pixel value for said center pixel.

3. A method as recited in claim 2, further comprising approximating an image from said averaged pixel values.

4. A peer group averaging method for enhancing images having n or more pixels, comprising:

identifying, at each pixel in an image, a group of nearby similar pixels and designating said nearby group as a peer group;

wherein said identifying of said group of similar pixels and designating of said nearby group as a peer comprising comprises

using a window diameter d and a peer group number n, selecting the peer group for a pixel from a window centered at the pixel, said peer group comprising the n pixels whose values are closest to the value of said center pixel;

averaging the values of the peer group pixels to obtain a new pixel value for said pixel; and

approximating an image from said averaged peer group pixel values.

5. A method as recited in claim 4, further comprising assigning said n similar pixels to said peer group based on a characteristic selected from the group of characteristics consisting essentially of intensity nearness to said pixel, velocity nearness to said pixel, color nearness to said pixel, and texture nearness to said pixel.

6. A peer group averaging method of image segmentation and enhancement, comprising:

for each pixel in an image, identifying a peer group of n similar pixels, wherein said peer group is selected from a window of diameter d centered at the pixel, and wherein said n similar pixels are assigned to said peer group based on a characteristic selected from the group of characteristics consisting essentially of intensity nearness to said pixel, velocity nearness to said pixel, color nearness to said pixel, and texture nearness to said pixel;

averaging said pixels and obtaining a new pixel value for said center pixel; and

approximating an image from said new pixel values.