Research Of Peripheral Items Detection And Classification Technology In Human Image

The most important items around human body are clothes,images of which are emerging inlarge numbers with the rising of online shopping. The research of clothing image classificationbased on context is becoming more and more popular recently.The main purpose of this study are as follows:(1) With studying of existing image segmentationtechniques, analyzing their principles, ranges of applications, advantages and dis advantages,combining with the properties of human image, to propose a suitable image segmentation algorithm.(2)By studying the existing variety of feature extraction program, combined with the maindifferences between the clothing objects, to find the best characteristics to describes different typesof clothes.(3)With analyzing the classification algorithms for different characteristics, training thefeature set of different objects, to get a classification model.For these purposes, a face detection based image segmentation algorithm using Grabcut wasproposed in this paper. The algorithm identified the possible pixels of clothes based on the results offace detection firstly, then removed the skin pixels by constructing the GMM of skin color and gotthe clothes object using the improved Grabcut algorithm, deleted the isolatedpoints of the dividedcontours lastly to complete the clothes object segmentation.After the segmentation of the clothes object, their color,shape,texture and location informationwas extracted, and the classification according to color, texture and shape was completed later. HSVcolor space was used to describe the overall color of the clothes;12features of texture informationwas extracted to train the SVM model, which is used for classification; seven different types weredivided into according to the different judgment process of upper and lower clothes;Freemanalgorithm was used to describe location information. Using the above algorithms, combined withthe unique nature of the clothing image, this paper proposes a human peripheral items detection andidentification system HODRS.With the test set of600different colors, textures, type with face clothes images, the test resultsto the achieved HIDCS were as following: face detection accuracy rate was88.9%, the textureclassification accuracy was90.58%, the average processing time of the image was2158ms, ofwhich face detection takes an average of636ms, image segmentation takes an average of845msand feature extractionIt takes an average of654ms.The test results show that the proposed algorithm can correctly detect and identify clothes object in images with people face, reached the preset acceptance criteria. The overall project hasbeen finished,and a related invention patent has been published.