Research On Image Thresholding Segmentation Algorithm Based On Lateral Inhibition Network

Image segmentation is the basic technique in image processing, image analysis and computer vision. It is widely used in the fields of military operation, industry inspection, clinical medicine and urban transportation. It has a high research significance with practical values. Among the numerous image segmentation algorithms, the thresholding is widely used due to its simple principle, intuitive processing and fast calculation speed.Two-dimensional(2-D) OTSU method is a typical thresholding segmentation method, but it always causes segmentation error when it processes the complex background image with the features of low contrast, uneven illumination, subtle discontinuity and strong noise. Considering that the lateral inhibition network has the functions of edge highlighting, contrast enhancement, discontinuity fitting and noise suppression, a novel 2-D OTSU method based on the lateral inhibition network is proposed in this theses in order to improve the original 2-D OTSU method on the contrast and illumination intensity adaptability, the capacity for fitting the discontinuities and the robustness to image noise. The 2-D cross entropy method can obtain a better segmentation result than the 2-D OTSU method when it is applied to the image with the large difference between the object variance and the background variance. A 2-D cross entropy method based on the lateral inhibition network is presented in this theses to deal with the complex background image that the difference between the object variance and the background variance is large. The major work of this theses includes the following two aspects:(1) In order to improve the 2-D OTSU method's adaptability to the complex background image, a 2-D OTSU thresholding segmentation method based on the lateral inhibition network is proposed in this theses. By utilizing the lateral inhibition network which is based on the human visual system to process the original image, the lateral inhibition image is obtained. A 2-D histogram with the gray information and the lateral inhibition information is established, on which the segmentation threshold can be calculated by using the maximum between-cluster variance criterion. Experimental results show that the proposed method can not only be well adapted to the contrast and illumination intensity, but also has the capacity for fitting the discontinuities compared with the OTSU method and the 2-D OTSU method. It has improved the robustness to image noise, too.(2) Aiming at the complex background image that the difference between the object variance and the background variance is large, a 2-D cross entropy thresholding segmentation method based on the lateral inhibition network is presented in this theses. For the establishment of the gray-lateral inhibition 2-D histogram, by maximizing the 2-D cross entropy criterion function, the segmentation threshold is obtained. Experimental results show that the presented method can be better adapted to the case in the complex background image that the difference between the object variance and the background variance is large, compared with the 2-D OTSU method and 2-D cross entropy method.