Font Size: a A A

Research On Shadow Detection Method For Single Outdoor Color Image

Posted on:2020-04-27Degree:MasterType:Thesis
Country:ChinaCandidate:W D RenFull Text:PDF
GTID:2428330599460543Subject:Engineering
Abstract/Summary:PDF Full Text Request
With the wide application of computer vision in industrial and agricultural production,astronomy,medicine and military science,it has attached much attention from more and more people.However,as a common optical phenomenon in real life,the existence of shadow brings the difficulties to computer vision tasks,such as object recognition,image segmentation,edge detection and so on.Therefore,effective detecting shadow becomes an urgent problem for the computer vision technologies.Through reading many literatures and comprehensively analyzing the domestic and international research status,by using machine learning and other relevant theoretical knowledge to conduct an in-depth research on the shadow detection method for the single outdoor color image.Firstly,the basic concepts of color image are introduced,and the machine learning knowledge involved is briefly introduced,including the principles of mean shift,random forest and convolutional neural network.Secondly,a shadow detection method based on random forest for a single outdoor color image is proposed.Firstly,the image is preprocessed.Secondly,the related features of shadow detection of a single image are analyzed.And the brightness feature,the color feature,the textural feature and the gradient directional feature are selected as the shadow detection features.Thirdly,the image is segmented with Mean Shift,the features are extracted from the images and normalized,input the shadow features into the random forest classifier to get pairs areas information.Finally,the shadow detection is completed according to the pairs areas information combined with the shadow features.And then,by analyzing the characteristics of single outdoor color image shadows,a convolutional neural network architecture is designed to realize shadow detection.Firstly,a network model structure based on global semantic information is established.Then the abundant masked shadow detection image samples are used to train the network model.Finally,the trained model is used to complete the shadow detection task for the single image.Finally,the two shadow detection methods for single outdoor color image proposed are verified through a series of comparative experiments,comparing to the existing shadow detection methods.
Keywords/Search Tags:shadow detection, random forest, convolutional neural network, global semantic information, shadow features
PDF Full Text Request
Related items