Research On Image Segmentation Algorithm Based On Improved Visual Attention Mechanism

With the further development of exploring, exploiting and utilizing the space resources, the future space combat will tend to increase and the earth outer space is gradually becoming a new territory for military struggle. In this situation, the space target recognition, pose measurement, information fusion and processing technology will be put more and more attention. Space-based measurement systems will be the future trend. How to obtain the target from the image getting from the measurement system is the basis and key point of the space target recognition and measurement. Of course, this research has important value and significance. So this dissertation takes a research on the method of digital image segmentation.Computational model for visual attention only considers the features of intensity, color and orientation, and it ignores the fact that the shape of target in image can also lead people to pay attention. So we introduce the edge feature which obtains through bilateral filtering algorithm including gradient weight. This dissertation uses nonlinear scale space because the linear may weaken some important image features, and uses a five-scale space to improve the efficiency of the algorithm. Taking into account the fact that people may put more attention to the scarce and central part of image, we revise the saliency map accordding to the salient area and the position of target; on the other hand, use the weighted strategy to merge feature saliency maps.This dissertation presents an approach for image segmentation based on visual attention mechanism. The improved visual attention mechanism is used to get the rough segmentation result, and then the active contour and the region growing based on contour are combined in order to obtain the accurate segmentation result. The image can be segmented automatically whether its background is simple or complicated. Experimental results show that the proposed method can overcome the problems occuring in the SVM algorithm, the adaptive threshold algorithm and the K-means clustering algorithm, that the segmentation result is not continuous and the earth background and object can’t be distinguished well. We can segment the complete and content target from the image with whether complex or simple background.In this dissertation, we study the image segmentation evaluation methods, and the objective evaluation methods are adopted which contains the goodness and the difference examining to evaluate the proposed image segmentation algorithm. The results of both methods prove that the proposed algorithm outperforms the other segmentation algorithms.