The Research And Realization Of The Similar Image Set Compression Based On Feature And Image Grouping

With the continuous emergence of new media applications in recent years,and the widespread availavility of high-quality image acquisition equipments,the scale of the image data explosively grows,which puts forward new requirements for the effective storage and transmission of image information.Image set compression is an effective method to solve this problem.Since any two images in a similar image set have a lot of similarities,if the redundant information could be effectively removed,we can increase the compression ratio of image set and save the storage space.In order to solve this problem,a similar image set compression scheme based on BoF feature and image grouping is proposed.The design,realization and performance of the compression scheme are also explored deeply.Firstly,based on the color information in the image,the color(HSV color space)SIFT feature descriptor HSVSift is proposed.The descriptor is used to train the feature dictionary.According to the feature dictionary and the HSVSift descriptor,the global feature of BoF(Bag of features)is constructed and is used to characterize the image.Secondly,we describe how to process BoF feature by using locality sensitive hash technique to compute index value.This index value is regarded as the fingerprint of an image or a BoF feature.We construct an index structure by using these index values.Besides,an image grouping algorithm which group similar image efficiently is proposed based on the structure.Thirdly,for the efficient compression of similar image sets,an image pair compression strategy is proposed.By using the hash index value,a similar image group is divided into some image pairs.One of the image pair is selected as the reference frame,and the other is regarded as the predicted frame.The reference frame is compressed by intra-coding,and the predicted frame inter-coded by against the reference frame.The experiments show that,in terms of ukbench image dataset,compression ratio of the similar image compression algorithms based on BoF feature and image grouping is 9.6% lower than Shi's scheme,but 13.6% higher than Zou's scheme.In addition,it is important that the average compression speed of our algorithm are 207 times and 6 times compared to competition techniques,and our algorithm supports fast access to the image,so our algorithm would be used for large-scale similar image set compression.