Image Quality Improvement Using Shift Variant and Shift Invariant Based Wavelet Transform Methods: A Novel Approach

Introduction

Over the decades, digital image plays an indispensable role in day-to-day life applications such as geographical information systems, astronomy, and satellite television. Images acquired by image sensors are generally contaminated, various factors responsible for affecting the quality of images, are imperfect instruments, data acquisition process, and interfering natural phenomena. Thus, de-noising is highly imperative for image data analyses which can be achieved by applying an efficient de-noising technique to compensate for image data corruption.

Many researchers have realized the importance of Multi-Scale Transform (MST) techniques in the field of image de-nosing which is more efficient compared to spatial domain based de-noising technique (Jain, 1989, Donoho & Johnston, 1994; Zhang & Wells, 2000; Chipman et al., 1997; Jansen, 2000). The shortcomings of the Weiner filter were pointed out and eliminated by researcher Donoho and John stone using Wavelet based de-noising technique (Jain, 1989, Donoho & Johnston, 1994). In spatial-scale domain based technique, first, an image is decomposed by applying MST on each source image, and then employ the optimal threshold condition to construct a composite multi-scale representation of the de-noised image. The de-noised image is recovered through an inverse MST. The commonly used MST techniques include the Laplacian Pyramid (LP) (Jain, 1989, Donoho & Johnston, 1994; Zhang & Wells, 2000) and DWT (Chipman et al., 1997; Jansen, 2000). Further, DWT has many advantages over LP methods in terms of localization and direction (Chipman et al., 1997; Jansen, 2000; Lang et al., 1995). Also, it is observed that DWT suffers from shift-variance and limited directionality. Thus, in order to overcome the limitation of DWT, the concept of SWT was introduced. The SWT is shift-invariant technique which is desirable in image analysis applications, such as, edge detection, contour characterization, and image de-nosing and image enhancement.

Thus, the main goal of this paper is to evaluate the performance of DWT and SWT techniques for the de-noising of images corrupted with different types of noises. The introductory knowledge of commonly occurring noises such as Gaussian noise, Poisson noise, Speckle noise, Salt and Pepper noise, etc. (Yang et al., 1995) are discussed in subsequent paragraph: