Analysis And Achievement Of An Image Classification System Based On Android

As mobile's computing becomes more powerful and its storage capability grows each year,bunch of pictures are captured and shared between users.It is necessary to organize and display these pictures more intelligent for users and mobile vendors.Traditionally,pictures are organized according to their folder names or capture date.However,because of the limited screen size,the drawbacks are increasingly obvious as the number of pictures becomes large.In this essay,the sources and properties of mobile pictures are analyzed the computing capabilities are discussed.Image classification is an important part of computer vision and faces many practical challenges.Extracted features based on image processing should distinguish different type of image correctly.Two available mobile classification models are based on supervised learning and unsupervised learning algorithms.Four image classifying methods are applied to classify phone's pictures more intelligently based on classical image classification algorithms and the human cognition theory.First of all,based on the raw information of image,a picture evaluation model is established to classify them into three categories: application cached,web downloaded and camera.Secondly,color histograms of picture are extracted in HSV color space to show the similarity of different pictures.Thirdly,a scene feature called spatial-envelope which is based on Gabor transform are applied to subdivide scene pictures.Two classical multi-category classifiers called KNN and SVM are trained to predict the real world scene images and the classification efficiency are evaluated.Finally,an object level feature called HAAR are extracted and trained by ADA boost algorithm to detect human faces within pictures.Besides,a friendly and intelligent gallery APP are designed and develpped to display above classification result on the android platform.The result shows that the classification mechanism proposed by this paper can describe the similarity of picture content and make up for the shortcomings of the existing gallery App.