Research On Spectral Image Segmentation Algorithm Based On Active Contour Model

In the field of image processing and computer vision,image segmentation is one of the most basic problems.In recent years,domestic and foreign scholars have deeply studied on the active contour segmentation algorithm.However,the existing methods lack universality and precision.It is necessary to carry on further exploration for the advanced segmentation algorithms for spectral images.On this account,this paper conducts corresponding research on the improvement of the level set equation in the active contour model and applies it to the infrared image and multi-spectral images segmentation.The paper mainly proposes and studies two novel active contour models:(1)An adaptive multi-feature segmentation(AMFS)model is proposed for infrared images.In this model,an adaptive weight coefficient to integrate the global and local signed pressure function is introduced to modify the level set formulation.The local term contains multiple statistical feature information and texture information by the Gaussian kernel.And it can denote the different areas comprehensively.Experiments are conducted with infrared images under different scenarios.Compared to traditional active contour models and edge detection method,the proposed model will not be affected by the background noise and it can deal with infrared image with inhomogeneous intensities,weak edges and low resolution.(2)A segmentation model based on the spatial and spectral information is proposed for multi-spectral images.The core of this model is the spatial-spectral SPF(SSSPF).On one hand,the value of principle components inside and outside the contour can be computed as the judgment criteria of evolution direction.On the other hand,the proposed model introduces weight coefficient to combine the two measures—spectral angel cosine and Euclidean distance(SAC-ED).The magnitude of driving force is determined by this integrated measure.Experiments demonstrate that.compared to other existing ACMs.the proposed model utilizes spectral information sufficiently and has a high segmentation precision.Compared to clustering methods,the proposed model will not be interrupted by details and obtains more distinct segmentation results for multispectral image.