طبقه بندي بافت بر اساس اطلاعات آماري تصوير

چكيده انگليسي :

Texture Classification based on Statistical information of texture image Maryam Mokhtari Esfidvajani Maryam mokhtari@ec iut ac ir Date of Submission Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan Iran Degree M Sc Language Farsi Supervisor Shadrokh Samavi samavi @cc iut ac ir Abstract Textures point to visual properties of objects surface which through them can determine the type of an object and its properties As a result the textures are important components of the world To this reason field to texture analysis has indicated in the computer vision In this field has studied texture classification segmentation synthesis and shape of texture Texture classification is one of the important issues in the field of texture analysis that its aim is allocation an unknown sample image to known texture classes Texture classification can be used to classify the rock surface wood identification face detection fabric classification and categorization of documents To evaluate the provided texture classification systems databases are made of texture images In this database images are designed that are almost the same as natural texture images and have their properties Natural images are often including variations of scale illumination rotation and viewpoints which in different texture classification methods have been tried to be covered more aspects of natural texture properties To this reason so far many techniques have been proposed that they can be classified in four group statistical structural model based and signal processing methods In every method extract different properties of texture image The extracted properties are studying with type of technique and texture images Today due to complexity methods and using synthetic properties it is problem to put them in a particular group because most of them are in several groups However in recent years studies show that between four categories algorithms based on statistical and signal processing techniques have provided powerful tools for texture classification This thesis introduces new methods for texture classification which most of them use the statistical methods Properties have been introduced for this purpose which is enhanced using them to describe the images accuracy of classifier system These properties are to illumination viewpoint and rotation invariant and can be extracted with little computation In this thesis the histogram of dominant gradient is presented which have been proposed two different methods for extracting this histogram This feature shows edge and orientation information and has proposed as a complement to commonly used local binary pattern Then in several experiments were tested different combinations of features to obtain features that are appropriate for texture images description and enhance system performance These properties are include color and type of local surface In some experiments has applied of used features in applications except texture classification that determined by different studies that this features are appropriate for texture description too Most of the proposed methods in this thesis could increase accurate of the texture classifier system in comparison with other existing methods Keywords Texture classification features of image Local Binary Pattern Histogram of Dominant Gradient color space PDF created with pdfFactory trial version www pdffactory com

مختاري اسفيدواجاني، مريم

طبقه بندي بافت بر اساس اطلاعات آماري تصوير

ويژگي تصوير , الگوي دودويي محلي , هيستوگرام جهت غالب , فضاي رنگ