Research And Implementation On Attractions Retrieval Based On Deep Hashing

In recent years,image retrieval has become one of hotspots of academic research which has been widely used in many fields such as industry,medicine,aviation and so on.Due to the ambiguity and complexity of image content,attractions retrieval has become a difficult problem in image retrieval.Image retrieval has received much attention since the emergence of constant local features and the groundbreaking work of Sivic and Zisserman based on Bag-of-Words.The retrieval system achieves a higher maturity by incorporating large visual codebooks,spatial tests and query extensions.These components constitute prior art for retrieval of specific objects.Another research is focused on compact image representation in order to reduce memory requirements and improve search efficiency.Representative methods include VLAD.Recent advances have shown that convolution neural networks are more advantageous in image retrieval.The description and feature extraction of the natural scene image are the most important in the retrieval process,but the traditional shallow features can't be retrieved because of the illumination,blur and size of the image.However,based on the deep learning,image retrieval has become the mainstream.Based on the previous research experience,this paper proposes a deep hash algorithm for searching the natural scenery with convolutional neural network and hashing.This paper firstly analyzes the present situation of content-based image retrieval at home and abroad,that is,the technique of image retrieval using visual features,and then summarizes these basic image features and points out the shortcomings of these methods.Then the hash algorithm is expounded,and the method is applied to the image retrieval,according to the experimental results show that the image retrieval time is reduced.Then,based on the knowledge of image retrieval based on the shallow features,a feature extraction algorithm based on depth learning is proposed,and the features are encoded by the hash algorithm to realize fast and accurate image retrieval of natural scene.In this paper,the main innovations are as follows: Firstly,a natural scenery image data set with label is established,and a depth network model is trained with the data set.Secondly,the model is based on the deep hashing algorithm and then applied to the natural scene retrieval system.Since the traditional hash method needs to use multiple features at the same time to achieve the performance of the deep hash method,the time to extract the feature seriously affects the coding speed in practical applications,based on the depth of the hash algorithm compared to improve the search performance in this paper.