Research On Image Reconstruction Algorithm Based On Quantum Optimization And Deep Learning

In the acquisition and transmission process of various images,phenomena such as low resolution,missing high-frequency details and blurred textures often occur,which affects the detection performance and efficiency of traffic monitoring,medical diagnosis,target tracking and other related video and image processing tasks.The current image reconstruction algorithm has poor anti-noise performance and low level of image detail restoration when faced with insufficient sampling data and prior knowledge.To solve the above problems,this paper studies and improves the reconstruction algorithm based on the quantum optimization theory,aiming at the effective restoration of extremely incomplete projected data and ultra-low resolution face images.(1)The research status of image reconstruction algorithm is investigated,and the basic theoretical knowledge of image reconstruction algorithm and quantum optimization algorithm is elaborated in detail.(2)A CT image reconstruction algorithm based on quantum optimization is proposed by combining quantum optimization algorithm with iterative solving algorithm for extremely incomplete projected data.Firstly,according to the actual situation of CT reconstruction,the qubit is initialized on the basis of pseudo inverse.Secondly,in order to enhance the ability of the algorithm to jump out of local optimal,a small limit is set and a new quantum gate is further designed.Finally,in order to overcome the shortcoming of the traditional iterative solution method,the iterative solution process and quantum evolution process are used alternately,and the comparison and experiment of different combination methods are carried out.Compared with other mainstream CT reconstruction algorithms,the proposed algorithm has excellent performance in image detail restoration and artificial artifact removal.(3)For ultra-low resolution face images,a deep network for face super-resolution reconstruction based on quantum optimization algorithm to adaptively determine weights is designed.The entire network structure is divided into a high-resolution face image generation network and a priori information estimation network: the main part of the high-resolution face image generation network is an improved generative adversarial network,which is suitable for face super-resolution reconstruction;face priori information estimation network is used to calculate the priori information of the geometric structure of the generated face,which provides structural information for the generation of high-resolution face images.Among them,A collaborative cycle mechanism is used to gradually enhance the priori of the image;and to simulate human visual attention mechanism,PSNR based on visual saliency mechanism and SSIM improved evaluation method are proposed.Experiments show that the algorithm proposed in this paper can reconstruct complete face information better than other deep learning-based algorithms.