An Approach to Image Compression by Using Sparse Approximation Technique

Abstract

Image compression has been a widely researched field for decades. Recently, there has been a growing interest in using basis selection algorithms for signal approximation and compression. Signal approximation using a linear combination of basis from an over-complete dictionary has proven to be an NP-hard problem. By selecting a smaller number of basis than the span of the signal, we achieve glossy compression in exchange for a small reconstruction error. For the past few decades orthogonal and biorthogonal complete dictionaries such as Discrete Cosine Transform (DCT) or wavelets were the dominant transform domain representations of signals. Over-complete dictionaries have been intensively studied and successfully applied to various application such as image de-noising, compression etc. The DCT and wavelet based compression methods suffer from blocking and ringing artefacts and also they are unable to capture the directional information. So, in this context an investigation has been made by using sparse coding method by Orthogonal Matching Pursuit (OMP) algorithm. In this proposed work, conventional DCT can be replaced by a set of trained dictionaries. For dictionary construction we have used a combination of DCT and Gabor basis and to encode the trained dictionary elements we have employed OMP algorithm. Experimental results demonstrate that the proposed method provides gains in Rate Distortion (RD) performance and improvements in perceptual quality.

The HTM Spatial Pooler—A Neocortical Algorithm for Online Sparse Distributed Coding

“…A commonly used criterion is how well one can reconstruct the inputs given the sparse activations and a set of learned basis vectors. Although these studies explain receptive structure in primary visual cortex and lead to practical machine learning algorithms for feature selection (Gui et al, 2016 ) and data compression (Pati et al, 2015 ), the purpose of neural computation is more than preserving information. In this paper, we take a different perspective and ask how computational properties of the HTM spatial pooler contribute to downstream cortical processing in the context of HTM systems.…”

The HTM Spatial Pooler – a neocortical algorithm for online sparse distributed coding

“…A commonly used criterion is how well one can reconstruct the inputs given the sparse activations and a set of learned basis vectors. Although these studies explain receptive structure in primary visual cortex and lead to practical machine learning algorithms for feature selection (Gui et al, 2016) and data compression (Pati et al, 2015), the purpose of neural computation is more than preserving information. In this paper, we take a different perspective and ask how computational properties of the HTM spatial pooler contribute to downstream cortical processing in the context of HTM systems.…”

Interpolation and denoising of seismic signals using orthogonal matching pursuit algorithm: An aplication in VSP and refraction data

“…Sparse signal representation in overcomplete dictionaries is an analytic tool that has been satisfactorily implemented in multiple signal and image processing applications such as image compression [15], audio denoising [16], and seismic signal pre-processing [17]. Its principle basically consists in representing the signal of interest as a linear combination of a few columns from a previously specified redundant matrix called dictionary, where each column of this matrix is called an atom of the dictionary.…”