Improvement Of Contour-Based Image Retrieval Using Pairwise Geometric Histogram

Because it is not simple to find the feature of image contour which fit the human eyes' sense, Contour-based image retrieval is a problem full of challenge in content-based retrieval techniques. The description of object contour is a difficult problem in it.This paper focus on improving the description methods of object contour in contour-based image retrieval. Therefore, we use two dimensional histograms that encode both the local and global geometric properties of the shapes. We compare with various description methods of object contour, and then introduce the pairwise geometric attributes which are the directed segment relative angle and directed relative position. Afterwards, we use the description method of object shapes in contour-based image retrieval. Besides, we also discuss these pairwise attributes which are not only robust to partial occlusion or loss of data, but also robust to variations in line description caused by fragmentation noise. They are also invariant to changes of scale, rotation and translation. We conclude that a relation histogram of pairwise segment attributes presents a very efficient way of indexing into large databases.