Image Compression using a New Adaptive Standard Deviation Thresholding Estimation at the Wavelet Details Subbands N.S.A.M Taujuddin Faculty of Electrical and Electronic Engineering, Universiti Tun Hussein Onn Malaysia, 86400 Parit Raja, Batu Pahat, Johor, Malaysia. shahidah@uthm.edu.my Rosziati Ibrahim Faculty of Computer Science and Information Technology, Universiti Tun Hussein Onn Malaysia, 86400 Parit Raja, Batu Pahat, Johor, Malaysia. rosziati@uthm.edu.my Suhaila Sari Faculty of Electrical and Electronic Engineering, Universiti Tun Hussein Onn Malaysia, 86400 Parit Raja, Batu Pahat, Johor, Malaysia. suhailas@uthm.edu.my Abstract—The process before quantization stage in compression process is a very crucial stage espeacially in application that require a high compression ratios. So, in this paper, we propose a new method of image compression that is based on reducing the wavelet coefficients in wavelet details subbands. It is based on the concept of local subband wavelet coefficients minimization to find the optimum threshold value for wavelet coefficients in each detail subbands. The proposed method decomposed the image into LL (low resolution approximate image), HL (intensity variation along column, horizontal edge), LH (intensity variation along row, vertical edge) and HH (intensity variation along diagonal). The coefficients in details subband retrived from the decomposition process is then manipulated in such a way that the nearly zero coefficient is discarded while the rest is remained. This process will reduce the unsignificant wavelet coefficient that leads to a great compression ratio while preserving the informative data to produce a good image quality as can be seen in the experiment done. Keywords— Wavelet Coefficients, Details subbands, Thresholding, Discrete Wavelet Transform (DWT) I. INTRODUCTION In recent years, digital image has rising popularity and has been becoming increasingly important. With a huge number of image application available online and mobile, it require a huge storage space that also burden the network capability [1]. Compressing an image is one of the promising solution that can reduce the amount of redundant data[2]. Besides, it will decrease the storage space, transmission time, bandwidth utilization and enable rapid browsing and retrieval of images from database [3], [4]. II. WAVELET IN IMAGE COMPRESSION Wavelet is one of the promising tools in image compression. There are three main properties of wavelets; (a) separability,scability and translability (b) multiresolution compatibility (c) orthogonallity With these properties, wavelet works well in image processing in such a way like the human vision do. It provides a much more precise analysis in digital image, movies and signal. It also has widely been used in data compression, fingerprint encoding and also image processing. Wavelet is a ‘small’ wave that zeroing the value outside the fixed interval time or space. It can be shifted or scaled to decompose a signal into different scales of resolution. Wavelet has special features in which it has flexible window size to determine accurately either time or frequency. It uses narrower window at high frequency for a better resolution while using wider window at low frequency for better frequency resolution. The behavior of wavelet has some similarity with the human DNA behavior: duplicating the essential unit, shift into different permutations providing a varied range different properties. Typically, wavelet uses two filters, namely analysis filter and synthesis filter. The analysis filter is used to split the original signal to several spectral components called subband. The signal is first will passed a low pass filter for approximation coefficients outputs that resulting a smooth effect. Then, it will passed through the high pass filter for the detail coefficients that enhance the details. In the analysis filter, some points need to be eliminated. This operation is called downsamping, usually illustrated as ↓2. The process is done to maximize the amount of necessitate details and ignoring ‘not-so-wanted’ details. Here, some coefficient value for pixel in image are discarded or set to zero. This is called as the thresholding process and it will give some smoothing effect to the image. The process of downsampling the image using DWT is illustrated as in Fig. 1. 109 ISBN: 978-1-4799-6210-5 ©2015 IEEE