Content-based Image Retrieval Technology Research Based On Object Extraction

Image contains large amount of information, intuition and easy to attract attention.With the rapid development of Internet technology today, quickly and accuratelyretrieval user required image from a large image database has become the focus of ourresearch. Due to the diversity of the image content, Content-based Image Retrieval(CBIR) technology is gradually being proposed and developed into the core researcharea of the image retrieval. The CBIR system is using the underlying feature of theimages to compare similarity in order to retrieve the images. To strengthen theeffective feature extraction of the images, this paper proposes an image retrievaltechnology based on object-extracted. Around the key technology in the content-basedimage retrieval, this paper makes a comprehensive study of the image objectextraction algorithm and retrieval algorithm. The mainly research includes t hefollowing aspects:(1) Image retrieval based on object extraction of gradient thresholdTo strengthen the effective feature extraction in the field of image retrieval, animage retrieval technology based on object-extracted is proposed. For grayscaleimages, this paper proposes an improved two-dimensional Otsu gradient thresholdfast iterative algorithm to segment image. This method uses the image gray levelinformation and the gray level-the maximum gradient two-dimensional histogramto select threshold, and use the iterative thought instead of exhaustive search tosegment object quickly at the same time. The shortcomings of segmentationinaccurate and the high computational complexity in the traditional two-dimensionalOtsu algorithm is overcome. The validity of this method has got verified by analyzingthe retrieval efficiency of extracted and matched the target object feature.(2) Image retrieval based on object extraction of graph cut algorithmIn object-based image retrieval technology, Grabcut image segmentationalgorithm is used in the Image retrieval system for color image. Grabcut algorithmcould effectively use the color information of the image, using the Gaussian mixturemodel to establish the color image data model instead of the histogram. This methodupgrades the image segmentation to the field of color, and uses iterative methodreplace one estimate the Gaussian mixture model parameter, improves the accuracy of segmentation. At the same time, it uses non-full label method to do the color markers,reduces user interaction workload. User simply needs draw a rectangle around thetarget area, then use of combinatorial optimization techniques to minimize the energyfunction to obtain the optimal segmentation. Add to Grabcut algorithm in the imageretrieval system, extracts and matches the image feature after completed segmentation.Experiments show that the algorithm can be satisfied with the retrieval results.(3) Implement an object-based image retrieval systemDesigning and implementing the object-based image retrieval system. Adding tothe image segmentation algorithm of this paper to extract object in the system. Theexperimental results demonstrate the effectiveness of the proposed method.