Research Of Face Detection And Face Super-resolution Reconstruction Based On Convolutional Neural Network

The purpose of computer vision is to realize the expression and understanding of the environment.In a broad sense,it is a discipline that "gives machine vision." By researching related theoretical technologies,intelligently extract information from high-dimensional data such as images,audio,and video,analyze and interpret the high-dimensional data content,and realize the ability of human-machine interaction.With the development of science and technology and the successful application of artificial intelligence in various fields,the research on computer vision has received more and more attention from scholars and experts.Applications such as target detection,target recognition,image reconstruction,and text recovery are covered in this area.This paper focuses on the two branches of object detection and image reconstruction in computer vision,and uses convolutional neural networks to implement face detection and face super-resolution reconstruction.As the most intuitive research and application direction of computer vision,face detection has benefited from technological advances and improved processor performance,and gradually the application of this technology product has appeared.The face detection algorithm based on convolutional neural network selects Wider-Face face data set for the experiment on the data set.Inspired by the Yolo algorithm,based on the structure of Yolov3-Tiny network,combined with the residual network to improve it.As an important part of the detection algorithm,Anchor uses the K-Means algorithm to generate Anchor values that are more in line with the face data set.After the down-sampling layer of Yolov3-Tiny,a residual network is added to avoid loss and loss of information transmission and improve the detection accuracy.The face detection algorithm in this paper mainly analyzes the accuracy and speed.Limited by actual application scenarios,such as: hardware equipment,external environment,etc.,the quality of the collected images is poor,and model trainingcannot be performed directly.The source of data information related to face detection is usually monitoring data,and the quality of information is greatly affected by hardware equipment.Therefore,the research on face super-resolution reconstruction technology has important practical significance.The face super-resolution reconstruction algorithm based on the convolutional neural network uses the CelebA data set for the experiment on the data set.According to the degree of influence of the different color channels of the RGB color space channel of the face image on the image distortion effect,through the experimental effect comparison,the three The channel color spaces are all reconstructed to avoid distortion effects caused by single-channel reconstruction.This paper takes SRCNN as the basic network structure,and improves the data preprocessing algorithm and network structure model.The Lanczos algorithm is an algorithm that transforms a symmetric matrix into a symmetric triangular matrix through orthogonal similarity transformation.It can be used for resampling and interpolation filtering.The Lanczos algorithm achieves better reconstruction results than the cubic interpolation algorithm.In the data preprocessing stage,Lanczos algorithm is used to replace the original cubic interpolation algorithm.The Inception V3 module was introduced into the SRCNN network structure.First,the feasibility of the experimental idea was verified in the MINIST data set,and the expected experimental effect was finally achieved.The PSNR of the face image reconstruction was 36.61.