Research Of Image Retrieval Based On Hashing Algorithm

Internet era, as well as widespread use of digital cameras and a variety of image processing tools on the Internet produced a flood of images, large-scale application of image data is facing a huge challenge, especially how to efficiently retrieve a sense of massive data similar image of interest. To solve the problem of large-scale image retrieval efficiency, widely used hashing-based image retrieval algorithms. However, its retrieval accuracy cannot make people satisfied, we propose to improve the retrieval efficiency improvements spectral hashing of image retrieval. The main work is as follows1. This dissertation introduces iterative quantization hashing’s idea to Spectral Hashing, which is minimize quantization error. This algorithm to get rid of Spectral Hashing’s assumption that uniformly distributed data requirements, and can keep the data in the original space similarity. In order to make this algorithm applied to different sets of image data, the introduction of machine learning Boosting algorithm to determine the threshold, so that the algorithm is more adaptive and more widely used. In this paper, the algorithm is in the public image data sets were verified, the experimental results show that this method is superior spectral hashing, locality sensitive hashing and iterative quantization hashing.2. Based on the above research, design and implement a prototype image retrieval system based on a hashing. The system provides visual retrieval services for hashing-based image retrieval’s research. Can be used as a hashing-based image retrieval algorithm platform to run the current mainstream hashing algorithm and obtain the necessary experimental data, the mean average accuracy, recall and precision rate. Realization of the platform not only facilitates research scholars related hashing-based image retrieval algorithm, and also provides image retrieval service for the average user. The modular design enables the system to easily add new hashing methods and other functions, the system has the scalability and adaptability.