Browse Prior Art Database

Fast Discrimination Between Homogeneous and Textured Regions

IP.com Disclosure Number: IPCOM000042515D
Original Publication Date: 1984-May-01
Included in the Prior Art Database: 2005-Feb-03
Document File: 2 page(s) / 15K

Publishing Venue

IBM

Related People

Dinstein, I: AUTHOR [+4]

Abstract

This invention relates to a method for fast discrimination between homogeneous and textured regions in digitized images. The novelty of the method is in the application of the operator as well as in its fast computation. Consider an operator which sets the center pel of a sub-image window to be the difference between the maximum and the minimum gray levels of pels within the window. This difference is high within textured regions and low within homogeneous regions. When the window size is small, say, 2x2 or 3x3, such operation can be used for edge detection. When the window size is selected to be larger than the considered texture particles, the operation can be used to discriminate between homogeneous and textured regions. Let F be a digital image function, and K any positive integer.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 48% of the total text.

Page 1 of 2

Fast Discrimination Between Homogeneous and Textured Regions

This invention relates to a method for fast discrimination between homogeneous and textured regions in digitized images. The novelty of the method is in the application of the operator as well as in its fast computation. Consider an operator which sets the center pel of a sub-image window to be the difference between the maximum and the minimum gray levels of pels within the window. This difference is high within textured regions and low within homogeneous regions. When the window size is small, say, 2x2 or 3x3, such operation can be used for edge detection. When the window size is selected to be larger than the considered texture particles, the operation can be used to discriminate between homogeneous and textured regions. Let F be a digital image function, and K any positive integer. Define W(F,K,i,j) to be the set

(Image Omitted)

where [I] is the floor function, which is the largest integer smaller than or equal to
I. Define an image MD(F,K) for any image F and positive integer K such that

(Image Omitted)

where MAX(S) and MIN(S) are the maximum and minimum, respectively, of all the pixel values in the set S. Hence, MD(F,K) is the difference between the highest and lowest gray levels within a KxK window. It is, of course, the maximal difference between the gray levels of any pair of pels belonging to that window. We refer to this operator as MAXDIF. The method utilizes the following property of maximum (minimum) values of sets. Let M1 and M2 be the maximum (minimum) values of the sets S1 and S2, respectively. Then, the maximum (minimum) value of M1 and M2 is also the maximum (minimum) value of the sets S1 union S2. Next, there is described the method in a formal manner. Let FXA(i), FXB(i), FYA(i), and FYB(i) be images, i=1,2,.... Let FXA(1) = F, and execute the following: For i :=1 to (log K) Translate FXA(i) by 2**(i-1) in the x-direction, and let the resulting image be FXB(i). Compute the MAX image of FXA(i) and FXB(i), and let the resulting image be FYA(i). Translate FYA(i) by 2**(i-1) in the y- direction, and let the resulting image be FYB(i). Compute the MAX image of FYA(i) and FYB(i), and let the resulting image be called FXA(i+1). end If K is a power of two, then by the end of log K loops there is obtained the image MD(F,K), slightly shifted. If K is not a power of two, then let [log K] = m, and N = 2**m; i.e., N O K O 2*N. Now, translate image FXA(m+1) by K-N pels in the x- direction, and denote the resulting image by FXB(m+1). Compute the MAX image of FXA(m+1) and FXB(m+1), and let the result be called FYA(m+1). Translate FYA(m+1) by K-N pels in the y-direction, and let the resulting image be called FYB(m+1). Compute the MAX image of FYA(m+1) and FYB(m+1), and denote the result by FMXDF. The obtained image, FMXDF, is the image MD(F,K), slightly shifted. More precisely, let [((K-1)/2)] = Ic, then, MD(F,K)(i,j) = FMXDF[ i+Ic-(N-1), j+Ic-(N-1)] where N is defined...