Image Processing Algorithms Based On Deep Convolutional Generative Adversarial Network And Quantum K-Nearest-Neighbor

An increasing number of image data is inevitably produced in people’s daily lives,since we entered the era of big data.Image,as the main form of big data,whose data scale has increased sharply.The research of image denoising and classification meets real challenges for the larger and larger dataset to be processed.One of the difficulties in image denoising and classification is that the calculation amount in the processing process is extremely large resulting in unsatisfactory image processing speed and processing effect,and it also consumes plenty of computing resources.Therefore,to solve this problem,this dissertation combines quantum computing with traditional image processing algorithms,and designs and experiments two image algorithms based on the deep convolutional generative adversarial network and the K-Nearest-Neighbor algorithm.The main research work is as follows:Enlightened by the idea of a zero-sum game in game theory,an image denoising algorithm of the improved deep convolutional generative adversarial network is proposed.The image denoising algorithm employs a denoising auto-encoder as the generator and a deep convolutional neural network as the discriminator.Through multiple confrontation training between the generator and the discriminator,it achieves the purpose of obtaining a better image denoising effect.The MNIST dataset is adopted for simulation experiments.The presented algorithm is compared with the BM3D image denoising algorithm and the denoising convolutional neural network algorithm.The experimental results reveal that the image processed by the presented image denoising algorithm not only has relatively excellent visual effects,but also has a better performance in the image quality evaluation index.Combining quantum computing theory,a quantum K-Nearest-Neighbor image classification algorithm based on K-L transform is proposed.The image features are extracted by the K-L transform.Then the image features are mapped into quantum states by quantum coding.Next,the Hamming distance between image features is computed and utilized to express the similarity of the image.Afterward,the image is classified by a new distance-weighted k value classification method.Finally,the classification result of the image are obtained by measuring the quantum state.Theoretical analysis shows that the time complexity of the presented quantum K-Nearest-Neighbor image classification algorithm is only O(?),which is significantly reduced compared with the classical K-Nearest-Neighbor algorithm.Simulation experiments based on MNIST,Fashion-MNIST and CIFAR-10 data sets demonstrate that the proposed quantum K-Nearest-Neighbor algorithm has relatively higher classification accuracy.