Statistical Methods Of Image Recognition Technology

With the aim of generating information of interest through data mining,the processing of images and videos has become a major branch of the Internet industry today.At present,the recognition technology for pictures or videos is at the peak of research.The purpose of this thesis is to design an image classification machine through image data to accurately identify different things in a picture and apply various processing methods to the video data to identify its background image.The specific research content are as follows:The thesis introduces the concept of Skew distribution and the establishment process of Gaussian Mixture Model.It combines the Steepest descent and EM algorithm to solve the numerical problem,builds the Logistic regression model and the Neural network model performs corresponding supervised learning,then use the improved k-means clustering algorithm to sort unsupervised.For the construction part of the image recognition machine,Firstly,use web crawler to obtain a large amount of image data and to improve clustering algorithm to make it have a memory.Secondly,cluster the identified pictures,separate a picture into different parts,then combine Principal Component Analysis and Convolutional Neural Networks to learn the pictures' data obtained by the reptile.At last,We can get the model to sort pictures into different types.In this thesis,video background are recognize by three different methods:Normal distribution hypothesis,Gaussian Mixture Model combined with EM,Partially normal distribution of different kernel functions.Compared and discussed the advantages and disadvantages of three methods in both stable background change and unstable background change.The data used by this thesis are image video data.The algorithms and models used in this thesis are implemented using Python program packaging.There are test progresses and test results after each model implementation and algorithm in this paper.The image test results show that the proposed method has certain practicability.