Image Decomposition Research Paper

IMAGE DECOMPOSITION TO DETERMINE UNDESIRABLE PATTERNS

G. Sangeetha, D. Madhina Banu

PG Scholar, Assistant Professor

Department of CSE, B.S Abdur Rahman University, Chennai.

[e-mail]

Abstract— Image is decomposed into multiple components has been an important research topic for many image processing applications such as image enhancement, denoising and inpainting. In this paper we present an image decomposition framework it is automatically decompose an input image. The proposed framework first learns a dictionary on the high spatial frequency part of a given image, based on the sparse representation for reconstruction purpose. Clustering is performed based on the observed dictionary output via affinity propagation which is

The purpose of image processing is to observe the image that is not visible, to create a better image, distinguish the object in an image and measure various objects in an image. Real time application of image processing is e.g., finger tip tracking. Image is a representation of external form of a thing. Image decomposition is to determine multiple source components and its weights [1]. For example an image is divided into texture and nontexture. Texture part conveys the global part of an image and nontexture part conveys the fain grain information. In our proposed method image decomposition frame work is created for de-noising application. Proposed method of image decomposition is done based on the dictionary learning. High spatial frequency image is used to done dictionary learning. Our proposed method of image decomposition framework uses two dimensional images. An image has width, height and doesn’t have depth it is known as two dimensional images. Image decomposition is used to remove the noise in an image. Noise is missing pixel values in the original image. Consider a fundamental problem of decomposing an image of N pixels into C different N-dimensional components, one need to solve a linear regression problem with N x C unknown variables. While this problem is ill-posed, image sparsity prior has been exploited to address

IH is obtained by subtracting smoothed version of IL from I. at this stage it is not clear how to identify the image components of IH which corresponds to undesirable patterns [1].

Traditional MCA approaches use a fixed dictionary like discrete cosine transform, wavelet, or curvelets basis to sparsely represent an image component. Challenging problem using MCA approach is to selecting a proper image set in advance for dictionary learning, it is not suitable for some real time applications. So different from traditional MCA approach, our proposed method the learning dictionary directly from the input image. After completing dictionary learning the remaining task is automatically done by our proposed framework such as removing undesirable patterns and image reconstruction [1], [2].

Context aware image component learning is done by based on the similarity between dictionary atoms. The number clusters is not known in before applying affinity propagation. After applying affinity propagation to group the atoms contains similar texture or edge information. The similarity function