A Fast Two Pass Multi-Value Segmentation Algorithm based on Connected Component Analysis

Abstract: Connected component analysis (CCA) has been heavily used to label binary images and classify segments. However, it has not been well-exploited to segment multi-valued natural images. This work proposes a novel multi-value segmentation algorithm that utilizes CCA to segment color images. A user defined distance measure is incorporated in the proposed modified CCA to identify and segment similar image regions. The raw output of the algorithm consists of distinctly labelled segmented regions. The proposed algorithm has a unique design architecture that provides several benefits: 1) it can be used to segment any multi-channel multi-valued image; 2) the distance measure/segmentation criteria can be application-specific and 3) an absolute linear-time implementation allows easy extension for real-time video segmentation. Experimental demonstrations of the aforesaid benefits are presented along with the comparison results on multiple datasets with current benchmark algorithms. A number of possible application areas are also identified and results on real-time video segmentation has been presented to show the promise of the proposed method.