Font Size: a A A

Image Classification And Retrieval Based On Merged Feature And Contour Feature

Posted on:2013-01-31Degree:MasterType:Thesis
Country:ChinaCandidate:J J LiuFull Text:PDF
GTID:2218330371989310Subject:Applied Mathematics
Abstract/Summary:PDF Full Text Request
With the development of science and technology, digital cameras have been widely used recently. As aresult, more and more images show on websites with the using of digital image processing technology.Providing users a good environment on-line is the basic requirement of sharing information. It means thatimage classification has broad application prospects; on the other hand, it is very important to give usersinformation they want quickly. So image retrieval has wide research value.Image classification and retrieval both work on feature extraction. Feature descriptors usually usedare low-level features such as texture and shape descriptors. Texture descriptor describes the spatialdistribution of image pixels. Shape descriptor represents the essential features of images. In this paper wedo two work respectively: one is giving a new feature descriptor on the on-line product imagesclassification; the other one is on image retrieval, where we give a new view point on image pre-processingand extracting feature based on similarity measure. The introduction is as follows.1. A local texture descriptor termed fan refined local binary pattern (FRLBP) was proposed. Then weconstruct an on-line classification of product images (OLCPI) system. Moreover, an OLCPI technique ispresented by fusing FRLBP feature and pyramid of histograms of orientation gradients (PHOG).Experimental results on a subset of product images database on Amazon and part of PI100havedemonstrated our proposed approach provides significant improvement as compared to the current existingmethods, with the highest classification precision increased by21%and the average classification timereduced by2/3. 2. To remove noise points among edge points this paper proposed a method getting the out-side edgeof objects. Then a shape feature based on out-side edge called HLDC is extracted with integration function.Experiments are done on MPEG7CE Shape-1Part B database. Results show that this retrieval isscale-invariant and rotation-invariant. Also it is effective on image retrieval.
Keywords/Search Tags:product image classification, FRLBP, PHOG, HLDC
PDF Full Text Request
Related items