Template quadtrees for representing region and line data present in binary images

Abstract

A template-based quadtree data structure for representing image features like regions and lines is described. This data structure called template quadtree (TQT) stores the region and line data present in a binary image in the leaf nodes if they correspond to one of the predetermined templates; otherwise it is quadrantized. This is recursively performed until the entire image is reduced to templates of different sizes in leaf nodes at different levels. The template size is 2^k × 2^k, where k is an integer greater than 0. The different types of templates considered are uniform color (black and white) horizontal, vertical, and diagonal lines. The number of templates possible for a given subimage of size 2^k × 2^k is 6 × 2^k. The region quadtree is a special case of the TQT is which the template corresponds to uniform color (black and white). Since the least size template is 2 × 2 the storage requirement of TQT is about four times less in the worst case situation like checkerboard. The representation of lines is based on pixels rather than storing lines that are fitted to the array of pixels. Thus this representation is accurate and reconstruction procedures are straightforward. The main feature of this representation scheme are: (i) it is capable of representing both region and line data; and (ii) it does not involve approximations. This paper describes TQT data structure, construction of TQTs from the binary images, and reconstruction. A brief description of some of the common operations on images using TQT data structure is given.