Browse Prior Art Database

VISUAL TEXTURE ANALYSIS, 2

IP.com Disclosure Number: IPCOM000128498D
Original Publication Date: 1970-Dec-31
Included in the Prior Art Database: 2005-Sep-16
Document File: 3 page(s) / 15K

Publishing Venue

Software Patent Institute

Related People

Azriel Rosenfeld: AUTHOR [+4]

Abstract

An algorithm for the detection of edges between re-gions that differ in average gray level is applied to the detection of other types of "texture edges", involving abrupt changes in coarseness ("visual cliffs") or direc-tionality. The algorithm is also modified to permit the detection of texture gradients, such as the coarseness gradients which arise when a surface is viewed in per-spective. The support of the U.S. Atomic Energy Commission, under Contract AT-(40-1)-3662, is gratefully acknowledged. The authors also express their appreciation to Roger Lipsett for his help in completing some of the work described here,

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 52% of the total text.

Page 1 of 3

THIS DOCUMENT IS AN APPROXIMATE REPRESENTATION OF THE ORIGINAL.

VISUAL TEXTURE ANALYSIS, 2

Azriel Rosenfeld and Mark Thurston

ABSTRACT

An algorithm for the detection of edges between re-gions that differ in average gray level is applied to the detection of other types of "texture edges", involving abrupt changes in coarseness ("visual cliffs") or direc-tionality. The algorithm is also modified to permit the detection of texture gradients, such as the coarseness gradients which arise when a surface is viewed in per-spective.

The support of the U.S. Atomic Energy Commission, under Contract AT-(40-1)-3662, is gratefully acknowledged. The authors also express their appreciation to Roger Lipsett for his help in completing some of the work described here,

1. Texture edge detection

If two adjacent regions on a picture differ sig-nificantly with respect to some textural property, they are often seen as separated by an "edge". A well-known example is the case in which the regions differ in "coarseness" (i.e., the elements which com-pose their textures have significantly different av-erage sizes); they may then give the appearance of a "visual cliff" (Figure la.) , in that the finer-textured region appears to be farther away from the viewer than the coarser-textured region. Differences in other tex-tural properties can also give rise to edges; for ex-ample, if the regions are streaked, and the predominant streak direction changes abruptly from one region to the other, the viewer may see a "directionality edge" (Figure 3a).

When adjacent regions differ in average gray level, the edge between them can be detected by computing dif-ferences between average gray levels taken over pairs of nonoverlapping neighborhoods at each point of the picture, and comparing the differences thus obtained for various sizes and orientations of the neighborhood pair and for nearby positions. Specifically [1], one can use square neighborhoods of sizes lxl, 2x2, 4x4, 8X8, 16xl6, and 32x32, and take differences of the av-erages taken over horizontally or vertically adjacent pairs of such neighborhoods at each point. The best horizontal edge size at each point is taken to be the largest size 2 h such that the next smaller size does streak detection algorithms [1] and comparing their ut-puts for various directions, In particular, a direction-ality edge can be converted into a gray level edge by applying streak detectors which are sensitive to the direction of the streaks on one side of the edge; this will yield an output picture having high average gray level on that side and lower average gray level on the other. The results of doing this, to Figure 3a, and then applying the gray level edge detection algorithm, are shown in Figure 3b-c.

2. Texture gradient d...