ROBUST REGION-BASED LINE DETECTION FROM POOR QUALITY IMAGES OF ALIGNED RECTANGULAR OBJECTS

摘要

A novel region-based weighted Hough Transform (HT) method for robust line detection in poor quality images of regular or rectilinear grids of rectangular objects is presented in this work. The proposed method decomposes a given binary image into connected regions, computes a rectangularity score for each region, filters out regions with low scores and, finally, uses a kernel to specify each region's contribution to the accumulator array based on the following two shape descriptors: a) its rectangularity, and b) the orientation of the major side of its minimum area bounding rectangle. Experiments performed on images of building facades taken under impaired visual conditions or with low accuracy sensors (e.g. thermal images) and comparisons between the proposed method and other HT algorithms, show an improved accuracy of our method in detecting lines and/or linear formations. Finally, in a document analysis application, the proposed method is used with success for skew detection and correction in rotated scanned documents.