Multi-Scale Adaptive Connected Component Labeling for Binary Images

Abstract

A multi-scale adaptive connected component labeling Algorithm was proposed. The image was first shrunk by levels of down-sampling, and the smallest image was labeled. Then the labeling result was propagated back to the original image level by level. With the design of joint memory structure, this algorithm needs no extra memory. During the backward propagation, double-layer decision tree was used to speed up searching. The backward propagation could be terminated earlier for application-specific constrains to fulfill application requirements