| With the rapid development of Internet and information industry, since 2012, people enter the era of big data, people are surrounded by a wide variety of information, such as text, audio, image, and video information. As a carrier of data, images give people a lot of information. And the number of images is growing fast.However, the resources are very limited for storing and transferring images. Because of that, how to represent the image efficiently becomes necessary. At the same time, images are noised inevitably, so there is a need to seeking a method of image denoising. Moreover, people aim to mine information from the image. Therefore, there is another problem to be solved, whether there is a certain algorithm can achieve recognizing image effectively or not.In order to solve above problem, this paper explores sparse representation theory, Deep Learning, Extreme Learning Machine.The main work is as follows:(1)Denoising deep extreme learning machines based on autoencoder(DDELM-AE) is proposed in this paper, it can be a newly developed feature representation method. And using this method can not only extract high level representation instead of the raw signals(e.g., images), but also remove the noise from the original image in this process,.(2)Based on the above method, the high level representation is taken as the input of the conventional K-SVD algorithm, which could boost the performance of the conventional K-SVD. Through the denoising “input”, the conventional K-SVD can get a denoising dictionary. Such a denoising dictionary is critical to K-SVD and dictionary learning.(3)In order to explain the generalization performance of the proposed method, the method is applied to four different data sets and every data set is typical in their fields. Especially in the multi-model data, the experimental results indicate the fact that our proposed method is very efficient in the sight of speed and accuracy. Consequently, it turns out that it has good generalization. |
PDF Full Text Request