Research On Infrared Image Segmentation Algorithm Based On Level Set

Compared with the natural image,the infrared image has the imaging advantage under extreme conditions.At the same time,imaging noise,complex background,and intensity inhomogeneity of target have brought great challenges to the image segmentation algorithm.Image segmentation algorithm based on Level Set represents high dimensional representation of low dimensional contour curves,and it can solve complex topological changes in image segmentation process,which has been widely concerned by scholars at home and abroad in recent years.However,the Level Set algorithm is not effective in segmenting infrared images with complex background and intensity inhomogeneity.Besides,the gradient descent method is used to minimize energy functional will lead to higher time complexity and it is prone to local optimal solutions.To solve the above problems,this paper carries out the research on infrared image segmentation based on Level Set,and obtained the following important results: the infrared pedestrian segmentation algorithm based on Intensity Adjustment Level Set Evolution,and the image segmentation algorithm based on global probability model of Level Set have been put forward.The content and innovation of this paper are as follows:(1)Performance of Infrared Ray acquisition of different infrared camera are difference,temperature of target in different locations is variance,or there is no significant difference in temperature between target and background,which will lead to fuzzy boundary and intensity inhomogeneity.Therefore,this paper proposes the target intensity adjustment Level Set evolution(Intensity Adjustment Level Set Evolution,IALSE)algorithm.Using the characteristics of the above infrared image imaging,the boundary enhancement processing is firstly carried out,that is,an improved Gaussian convolution kernel is constructed to extract more local neighborhood pixel information and a soft threshold mask is defined.Then a weighting equation is established to adaptively adjust the target intensity,reduce the intensity inhomogeneity,and obtain a more robust boundary indication function.Comparative experiments show that the algorithm can obtain more accurate infrared image segmentation results under the framework of Level Set.(2)In view of the infrared imaging characteristics of(1)and the use of gradient descent algorithm,which leads to the higher time complexity of the algorithm and is prone to the local optimal solution,the Global Probability Model Level Set Evolution(GPMLSE)algorithm is proposed.Based on the thermal diffusion physical phenomena described by the heat equation,the probability model is established.Probability is the heating value of pixels,which includes the distance relationship between the pixel and the pixels being set as heat source,and the global distribution information of the pixel with the remaining pixels.The inverse cosine function is used to map the probability model into a Level Set energy constraint term.In addition,the bias field model extracts local neighborhood information of pixels to form an energy constraint item,which fits the variance of the target intensity.Finally,the combined optimization algorithm is utilized to minimize the Level Set energy functional in the graph model,and the time complexity of the algorithm is reduced significantly,the global optimal solution is also obtained.Many group of comparison experiment results show that the algorithm can effectively improve the accuracy of infrared image segmentation with intensity inhomogeneity and complex background.The segmentation results do not depend on the location of the initialization contour curve and effectively resist noise interference.The extended experiments show that the algorithm can also effectively segment the nature image.