Research On Algorithms For Image Segmentation Based On Otsu Theory

Image segmentation is a key and important step of computer vision. It plays anessential part in subsequent task, suck as image analysis, feature extraction and targetrecognition. The purpose of image segmentation is to divide the image into somedifferent subimages, which are not overlapped. Image segmentation could make imagesbetter to be understood and analyzed by changing the representation of images. Afterseveral decades of development, there have been lots of different types of imagesegmentation algorithms. Among these algorithms, people pay much attention onthresholding methods that is simple and practical.This paper is main focusing on the subject:“The research on algorithms for imagesegmentation based on Otsu theory”. In order to simplify the computational complexityand improve the segmentation effect, this paper has proposed three correspondingimproved algorithms. The effectiveness of each algorithm has been proved by relevantexperiments. This paper’s main work can be listed as follows:1. Aiming at the problem that the traditional two-dimensional Otsu imagesegmentation algorithm can not be effectively used in the image segmentation while theimage contains lots of noise, an improved two-dimensional Otsu image segmentationalgorithm is presented in this paper. Firstly, an improved adaptive weighted medianfilter is used to filter noise image in order to eliminate interference of the noise. Then,the region division in two-dimensional histogram is transferred from four parties intotwo parties, and the vertical distance between halving line and original point is replacedof the two-tuples in two-dimensional histogram. This method reduces the spatialdimensions of searching threshold from two-dimension to one-dimension. As a result,this approach is greatly simplifies the calculation, and it is to correct the errors of followblindly ignore certain information. The experimental results of some images indicatethat the proposed algorithm has better segmentation effect.2. Aiming at the weakness of the computational complexity and huge calculation ofthe three-dimensional Otsu, a three-dimensional Otsu image segmentation algorithmbased on decomposition is presented in this paper. Firstly, this paper reduces thecomputational complexity and simplifies the calculation by decomposition,decomposing the original three-dimensional Otsu algorithm into three one-dimensionalOtsu algorithms. Then, after analyzing one-dimensional Otsu algorithm, this paper defines between-class distance and within-class distance, a novel algorithm with a newthreshold recognition function is also proposed, which considers the different roles ofbetween-class distance and within-class distance. The simulation results also prove theeffectiveness of the algorithm3. The three-dimensional Otsu have some drawbacks in contour details andobjective edge when segments images. To solve this problem, this paper first replacedthe mid-value by gradient, which is got by Sobel operator. Then we build a bridge tocombine between-class distance and within-class distance.Here the maximum scatterdifference discriminant criterion is the bridge.The experimental results indicate that thesegmented image’s contour details and objective edge are clear and easy to observe.