Research On Image Retrieval Algorithm Based On Spatial Distribution Entropy To Improve VLAD

With the rapid development of computer technology and multimedia technology,the digital image has become an important means for people to communicate.Digital images contain rich content,which can satisfy people's pursuit of information diversification.As a result,the number of images is growing at the rate of hundreds of millions per day.How to obtain images from uncounted images has become a major problem in the computer field,and image retrieval technology has been born.The original image retrieval technology is text-based.Text-based image retrieval requires artificial tagging of images,which has subjective ambiguity and cost of manpower,so it is difficult to meet the needs of retrieval in such a huge image database.Content-based image retrieval is to extract features from the image itself and describe the image in a way that is close to the human visual sense.These features describe the image more comprehensively,and can be better used in image matching and retrieval.Therefore,content-based image retrieval has become a hot research topic in recent years.In this paper,the image retrieval technology is studied to improve the ability of VLAD feature retrieval,and mainly complete the following work:1.SIFT(Scale-invariant feature transform)is one of the most commonly used image features.The most classic feature is the BOF(Bag of features).But the BOF algorithm has lost a lot of information when it is transformed into a visual word,so its retrieval ability has declined.VLAD(Vector of locally aggregated descriptors)features retain most of the information on the basis of BOF algorithm,so its retrieval ability is better than BOF algorithm.In this paper,the above three algorithms are studied carefully and the three are implemented.2.VLAD algorithm is a hot research topic in recent years,a large number of scholars have studied and improved the VLAD algorithm.In this paper,a variety of improved methods for VLAD algorithm in recent years are counted.The improved methods are classified,and their advantages and disadvantages are summarized.3.Entropy is one of the parameters that characterize the state of matter in thermodynamics,and its physical meaning is the measure of the degree of chaos in the system.In the field of image processing,this concept has been successfully used many times.Through the study of the VLAD algorithm,it is found that the VLAD features don't have the spatial location information of local features.This situation may lead to decline the retrieval capability of VLAD feature.In this paper,the concept of improving VLAD based on spatial distribution entropy is proposed.By extracting the location information of SIFT descriptors,the concept of entropy is used to describe the spatial distribution information of local descriptors belonging to a cluster,and the spatial distribution entropy is added to the traditional VLAD features.A large number of experimental results show that the addition of spatial distribution entropy can significantly improve the retrieval ability of VLAD features.4.By combining a variety of improved methods for VLAD features,The validity of the VLAD features added to the spatial distribution entropy is verified.The experimental results show that the retrieval ability of VLAD features can be further improved by adding a variety of improved methods based on the spatial distribution entropy.