Caption segmentation method in videos using isodata clustering of topographical features

Conventional caption extraction approaches often use a spatial variance method or a color reduction method such as color histogram and color quantization. These methods are time-consuming and restrict the colors for extracting captions. In this paper we propose a method that extracts caption areas using TFs (topographical features) and isodata color clustering. After extracting TFs points using the topographical characteristics of characters, the isodata clustering method is used to cluster the similar colors. The point-line-region method is used to construct the blocks for each of the clustered groups. Constructed blocks are verified and candidate areas are decided. Caption regions are determined by testing several conditions and verifying those candidate areas. Experimental results show that the candidate region extraction rate is 100%, and the character region extraction rate is 98%.