A Modified Image Filtering Method Based On LLE

Digital images collected by image sensors are generally contaminated in the process of data acquisition, digitization, compression and transmission. Many types of noises or degradations may be included in the image. Thus, it is necessary to take image filtering before image data is analyzed. Image filtering methods are generally categorized into two basic approaches: one is spatial domain approach, which is based on window and filters the image directly, such as mean filter, median filter, while another filtering approach, frequency domain method, is based on some transformation, such as Fourier and/or Wavelet transformation.Image filtering method based on Locally Linear Embedding (LLE) is a novel image filtering approach which is adaptive to all the noise types. In this paper we present the idea of the LLE filtering method firstly, after analyzing the problems of it, we improve its performance in two aspects. Based on the idea of the spatial domain approach, we use mask, which is calculated by LLE, to process the image pixel by pixel in our algorithm. Experiments demonstrate that it overcomes the block edge phenomenon of original algorithm and performs well when the image is contaminated by small noise. Considering the manifold learning character of LLE, we give another improved method based on image segmentation. In this algorithm, we use the reconstruction procedure in the segmented region, which is obtained by image segmentation. Many experiments demonstrate that compared with the traditional image filtering method, our algorithm has better filtering performance and keeps the adaptation character of original algorithm at the same time.