Study On Enhancement And Segmentation Algorithms For Undersea Hydrothermal Vent Images

The 21^st century is featured by the exploitation of vast ocean. Strong emphasis has been put on ocean exploration in order to fulfill the needs of Ocean Economic Development. Automatic object identification and recognition is of key importance in the development of Intelligent Undersea Vision System. Both research and undersea exploration indicate that the movement of undersea volcano and hydrothermal vent has close relationship with the distribution of undersea metal mineral resources and living creature activity. Under this background, this paper did some research on undersea color image enhancement and adaptive segmentation algorithm: 1. Under sea color image enhancement. We discussed the implementation of HIS and CIE Lab model in undersea color image, analyzed the principle of contrast enhancement using non-linear transform function and the high-frequency information enhancement using saturation feedback. Basing on the analysis of the undersea image features, low contrast, uneven lighting, blur texture details to name a few, this paper developed a hue preserved undersea hydrothermal vent image enhancement algorithm. This algorithm can produce higher contrast and better texture details. 2. Adaptive segmentation algorithm for undersea hydrothermal vent image. This paper carried out in-depth analysis of the features of three kinds of objects in undersea hydrothermal vent image: sea water, smoke and rock, discussed the strategy of segmenting the image step by step. Basing on this analysis, this paper developed an adaptive algorithm for undersea hydrothermal vent image. This algorithm carries a first step segmentation by thresholding basing on histogram analysis and a second step segmentation by local iterative fuzzy clustering. This algorithm has good adaptability for complex undersea environment. 3. Connected Component Detection on Binary Images. In order to select the major smoke area, connected component detection should be carried out on segmentation result. Basing on the analysis on traditional connected component detection algorithm, this paper developed a new connected component detection algorithm basing on image plane scanning. The algorithm scans the image pixel by pixel and records the local connecting relationship by pixel label, while at the mean time, the connecting relationship of entire image is record and mergered using dynamic linked list. This algorithm can detect the connected component both quickly and steadily.