The Image Content Retrieval System Based On Android

In recent years, with the rapid development of multimedia and Internet, mobile phones,tablets and other intelligent handheld mobile terminal rapid popularization, the way people livegreat changes have taken place. In the same time, various styles of image information is growingsharply, and traditional text-based image retrieval has can’t satisfy people’s need, content-basedimage retrieval technology arises at the historic moment, but how to real-time field in a largenumber of images quickly retrieve a user want to image information is still urgently needs tosolve the problem. Based on mobile computing platforms and content-based image retrievaltechniques are discussing in this paper. Used a strategy: Store and manage large amounts ofimage data use a PC, use a lower machine self-management of local small amount of imageinformation, and establish a connection with PC via3G or WIFI to achieve remote imageretrieval. Designed and realized the HSV block color histogram and gray level co-occurrencematrix of image retrieval.This system upper machine(PC) development based on Web Service of SSH framework toachieve a separation of the presentation layer, business logic layer and data services to improvethe performance of the program expansion, and create dynamic web pages quickly throughAjax2technology, the web of asynchronous updates, thus optimizing the website update speed.Lower machine based on the Android platform, utilizing Android graphical layout, data storage,networking and communication technologies to achieve a local and remote image retrievalsystem. In the extraction of low-level features of image, the color feature extraction algorithm ofHSV based on block color histogram and texture segmentation algorithm based on gray levelco-occurrence matrix, and based on the above two algorithms, put forward a comprehensivesearch strategy combined with HSV block color histogram and gray level co-occurrence matrix.Gray level co-occurrence matrix using the contrast of the image, the energy, entropy,correlation of the texture feature vector, and calculating the similarity image with the sampleimage library through the Manhattan distance, by comparing the threshold value, filtering theimage with high similarity, will feature images selected HSV color values again throughManhattan distance similarity matching, arriving at the searching results. In the relevancefeedback, the feedback approach taken to move the point of the relevant query using the userback to the positive and negative feedback of an image information, the query vector is adjustedto move in the direction of the positive cases, the direction away from the counter-examples tofurther improve the accuracy of retrieval.In image retrieval, not only the user can query the local real-time field image information, but also real-time query remote field image information via3G or WIFI feature. Before thesystem operation, the image preprocessing first database, the article takes the median filter, andthen extracting the processed image color and texture features, and stores the extracted featurevectors into the database. Retrieve images, users need to select the sample image to be retrieved,and to achieve the image pre-processing and feature extraction operation, if the user selects alocal image retrieval, the feature vectors and local feature library system will extract the featurevector comparison. In descending order according to the level of similarity, for finding images inthe order that the local library in the corresponding image, shown in the UI bit machine interface;If the user selects a remote image retrieval, then by Soap agreement to the extraction the featurevector is transferred to the host computer, the host computer via similarity detection algorithm,send up the feature vector and library feature vectors are compared sort, and then retrieve theresults transmitted to the next crew, by the next crew UI interface for users browse operation.Users can take advantage of the search results relevance feedback technique to retrieve theimages labeled as positive examples or counter-example image retrieval several times to meetthe retrieval needs.During the test, the use of64-bit Windows7system, of PC as, the upper computer, LenovoLe Pad as the lower computer, and the PC and the next crew to communicate via WIFI, andselect the Corel image library of1000as the experimental image simulation, the image size is256x384, resolution of72dpi. In the image retrieval process, flower, uses, dinosaurs and lacesmade these four images retrieval accuracy statistics respectively, the search results show that thehigher the image of dinosaurs such simple structure retrieval accuracy, which is based on HSVblock color histogram algorithm recall rate of46.3%precision rate of92%; GLCM algorithmbased on the recall rate was33.3%,66.7%precision; based on HSV and integrated algorithmGLCM recall rate of47.3%,94.7precision. Such structures on the lower palace complex imageretrieval accuracy, which is based on HSV color histogram algorithm block the recall rate of20%,precision of40%; based on GLCM algorithm recall rate13.3%precision rate was26.7%; HSV-based synthesis algorithm and GLCM recall rate was20.7%, the precision rate of41.3. Tofurther validate the feasibility of content-based image retrieval based on Android mobileterminals, its retrieval speed were tested, the test results show that the1000’s image library, thesame image in the remote retrieval speed is about7.8seconds, the image cached locally, theretrieval time is used again to2.1seconds, based on local image retrieval process is about2.3seconds. In the Retrieval process, the system is stable, retrieval speed, high accuracy, thefeasibility of content-based image retrieval in the Android mobile terminal implementation andhope for lower mobile terminal application and promotion of image retrieval technology can playa role in promoting.