A New Non-convex Sparse Coding Algorithm For Image Classification

Image classification techniques are significant to satisfy different people's requirements in processing image.A mass of images can be automatically classified by computers using image classification techniques,which has very extensive application in research,business and our daily life.Physiologists found that human's vision system is a procedure of abstraction of visual signal.As one of most important step,sparse coding gradually has an important influence on image classification after being introduced into the image classification technology.As a part of this thesis,we learn a famous model in image classification--linear spatial pyramid matching model and design an improved image classification method based on sparse coding,which is a non-negative model and non-convex sparse coding algorithm.Its characteristic is to use two properties of images.The remarkable features of images is sparse and similar features should be similar expressions.The algorithm seeks better solution through iterations in solving optimization problems.We apply this kind algorithm to image classification.Through the numerical experiment,we find that this new kind algorithm can preferably settle image classification problem,and the results are competitive to this kind methods.We also did some exploration in banknote recognition,and proved the significance of dictionary in sparse representation.We know that this method is very potential in banknote recognition.