Fast multi-level connected component labeling for large-scale images

Connected component (CC) labeling is time consuming during image segmentation and object identification, especially when input images are large-scale and binary converting uses multi-threshold. So in this paper, we present a fast multi-level CC labeling by combining online threshold segmentation and a one-pass labeling algorithm, also a detailed architecture for hardware implementation is proposed. During labeling, firstly a gray image is converted to several binary data flows with different thresholds. Then they are marked by run-length code respectively with 2× 2 scanning windows, in the meantime, the equal labels between two adjacent rows are obtained by these windows for CC merging. By executing this process until the end of input image, the CC labels are obtained after one-pass image scanning. Our method is tested on more than 1000 gray images which resolution are 1920×1080. Experimental results show that our algorithm can extract the positions and areas of CC in multi-threshold binary images at least 50 frames per second (fps) on a Stratix IV FPGA platform running at 109.7 MHz.