Research And Application On Image Emotion Semantic With Fuzzy Nerual Network

In today's society,the development of information science and technology has entered a booming stage,more and more important information is embodied in the form of text information,and it is also contained in images.In order to obtain the corresponding information from relevant media in time,people have developed many methods to accurately obtain arbitrary image information.Taking image as the focus of discussion,this thesis mainly studies the relationship between image content and its corresponding emotion,aiming to solve the problem of "semantic gap" caused by traditional content-based retrieval of digital image.According to the retrieval of the emotional semantic of image to achieve the purpose of conveniently querying the images required by the user.In the emotion-based image semantic mapping module,the neural network algorithms are used more by relevant scholars,but the traditional neural network algorithm have a poor generalization ability and a slow convergence speed.In this thesis,fuzzy neural network algorithm is used to map the emotional semantics of images,and neural network and fuzzy logic are combined to make up for the shortcomings of fuzzy logic in self-learning and the inefficiency of neural network in dealing with fuzzy problems.At the same time,genetic algorithm and ant colony algorithm are introduced to optimize the fuzzy neural network,so that the accuracy of the mapping is improved to a certain extent.The main research work and contents of this thesis are as follows:(1)The color and shape features of the digital image are fuzzified as the input of the fuzzy neural network,at the output of the network,the images of different emotion categories are labeled to start the training process.After the training process is completed and get the the classification model,the color and shape features of the image to be mapped are input,and the corresponding emotion feature vector is obtained after the defuzzification at the output of the model to realize the emotional semantic mapping of the image.(2)In this thesis,genetic algorithm and ant colony algorithm are used to optimize the fuzzy neural network respectively to improve the accuracy of mapping.Genetic algorithm is binary coding the weights and thresholds of fuzzy neural network as the initial individual of the optimization process,and substitute itinto the fitness function threshold in the experiment.The ant colony algorithm sets the weight and threshold of the fuzzy neural network to a corresponding set of multiple random non-zero numbers,In the process of searching for the best path from the ant colony to the food source,The ants find the best weights and thresholds from each set and substitute them into the fuzzy neural network to improve the accuracy of the emotional semantic mapping.(3)In the image retrieval module,this thesis proposes a sequential similarity detection algorithm to match images,The idea of the algorithm is to find the best sub-template of the image to be matched that matches the image in the data set to improve the accuracy of the retrieval.In the experiment,the color auto correlogram and the color fuzzy correlogram in the color correlogram were introduced as comparative experiments to compare the advantages and disadvantages of the retrieval effects using different methods.