The Nonlinear Feature Extraction Of Image And Its Application In Image Segmentation

Image recognition is a relatively new subject,because the scope of application is broad,medical imaging,weather maps,tidal observation,product appearance detection,face recognition and so on,are needed,so from the emergence of attracted widespread attention based image recognition is the image segmentation.This topic by analyzing and comparing a variety of image segmentation method,is applied to image segmentation of semiconductor devices,innovative use of nonlinear feature extraction in semiconductor image feature extraction and image segmentation,achieved good effect,for further practical application laid a good foundation.Texture segmentation is an important and elementary issue in objects recognition,image understanding and computer vision.It has made great development during the past few decades by the efforts of domestic and overseas researchers.Now,more and more different segmentation methods hAverage come into forth,and some of them can get satisfied results.However,they hAverage the usual disadvantages,such as less robustness,dissatisfied unsupervised results,and expensive consume in time and space etc.So far,most of the algorithms have been concentrated in the linear field.This paper is innovative in extracting the nonlinear characteristics for the positioning and detection of semiconductor components,which has a good effect.Hereupon,this thesis addresses segmenting texture images in nonlinear field.Base on the description complexity of texture images,it uncovers the chaos features of texture images,proposes a new algorithm which based on complexity measure to draw characters from texture images,introduces two segmentation methods that are unsupervised and supervised segmentation methods,then using them to segment texture images and compare the results,expand the research view of texture segmentation.The main contributions of this thesis are summarized as follows:(1)Different texture segmentation methods are studied and compared and compared.The advantages and disadvantages of these methods are summarized,and the applicability range,ability and existing problems are discussed.(2)The theory of complexity is introduced after the discussion of nonlinear science.The complexity of texture image is described and the chaos features of texture images are uncovered.(3)Large numbers of Brodatz texture and Uni-Bonn texture images and Image of semiconductor components are dealt with feature drawing by the means of complexity measure.Including 2-valued pre-procession with moment invariant,scanning images using Hilbert curve to realizing the reconstruction of image information,separating texture image into different blocks to sample complexity features etc.The experiment results proved that this description to texture structure with complexity has better clustering results.At the same time,the 2-valued pre-procession based on moment invariant which chooses threshold value by making full use of its character has high precision and accords with practical requirements.The Hilbert curve which can preserve the space relativity in scanning image realizes the reconstruction of image information.(4)Two different kinds of clustering ways that Fuzzy C-Means(FCM)technology and Support Vector Machine(SVM)technology hAverage been adopted to texture images' feature segmentation,and hAverage got satisfactory results.Meanwhile,the complexity measure has been compared with the common means.The experiments show that,on the condition of preserving a certain segmentation precision,the method based on complexity measure has least operation amount and highest time-efficiency.(5)Process of semiconductor has also been introduced.The locating algorithm adopts the complexity measure which we discussed before to describe the semiconductor structure,and to locate it.The tests show that the segmentation result is satisfied.(6)the paper introduces the design of the semiconductor device location identification system process,the components of the complexity of the algorithm is adopted in this paper,the front is introduced to measure the image texture description of the structure,and combining with the characteristics of the semiconductor device itself segmentation positioning,proved has better location performance.