Font Size: a A A

Study Of Convolutional Neural Network Applied On Image Recognition

Posted on:2013-01-18Degree:MasterType:Thesis
Country:ChinaCandidate:K XuFull Text:PDF
GTID:2248330395489225Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Artificial Neural Network (ANN), which is one of the many Artificial Intelligence technologies that inspired from the functioning of human brain, after its rapid growing at the later20th century, entered a stage of slow development. Deep Learning, inspired from the recent discovery of biology and neurology on animal and human optical nerve systems, by simulating the hierarchical structure of the human vision system, developed a deep network architecture which has hierarchical structure. Convolutional Neural Networks (CNN) is a technology that combines ANN and recent Deep Learning method, which is characterized by local receptive field, hierarchical structure, global learning for feature extraction and classification, has been applied to many image recognition tasks. Modern image recognition tasks focuses on the systems that can adapt to different kinds of recognition problems. Deep Network especially its special example CNN have drawn much attention recently, thus the research on apply the CNN on different type of image recognition tasks has great application value.In this paper, we first summarize the latest achievement on ANN and CNN fields, and give a thoroughly introduction on the basic concepts of ANN and CNN, including the algorithm that we will use in the work. Then we try to adapt the CNN to handwritten digits recognition and face recognition tasks, the main works in the paper list as follow:1) We constructed some CNN models with different number of feature extractors based on a classic CNN model and applied them on two distinguishing image recognition tasks, handwritten digits recognition and face recognition.2) By comparing the experimental result of Misclassification Rate on both handwritten digits recognition and face recognition with other existing methods, we proved that CNN can be applied to solve both recognition problem and achieved considerable good results and then give a belief analyze on the relative merits of the CNN model.3) Based on the experimental results obtained from those CNN models that have different structure, we analyzed those CNN models by comparing them with aspect of train speed and misclassification rate.By conduct these experiments, we proved that CNN can be used on handwritten digits recognition and face recognition problems without much modification and achieved good classification results, and then we analyze the influence of the number of feature extractors of each layers on the training and classification results.
Keywords/Search Tags:Convolutional Neural Network, Artificial Neural Network, ImageRecognition, Handwritten Digits Recognition, Face Recognition
PDF Full Text Request
Related items