Adaptive Image Segmentation Based on Fast Thresholding and Image Merging

Abstract

Image segmentation is the first essential and important step of low level vision. This paper proposes a novel algorithm for adaptive image segmentation, based on thresholding technique and segments merging according to their characteristics combine with spatial position. Our earlier work of getting the entire information of the histogram could help choose the multiple thresholds. However, not all the peaks of the histogram correspond to obvious structural unit in the image. Spatial information must be involved. This paper also suggests subjoining segments matching for video image tracking. They will give great help to image segmentation. The proposed algorithm can meet the real-time requirement and lead to higher segmentation accuracy, some types of texture can also be segmented well; it can be applied in many conditions, including complex target segmented. We describe the algorithm in detail and perform simulation experiments. The computation based on pixels can fully parallel processing to save time