Article Preview
TopIntroduction
With the rapid development of multimedia technology over the mobile devices on the Internet and broad-band communication networks, more image data are in need to be transferred over tighter bandwidth. This drives to the development of more advanced compression technology for image coding. Inevitably, images suffer a variety level of distortions in different compression coding processes (Wallace, 2003; Taubman & Marcellin, 2002). How to evaluate these images and determine which coder is better or more appropriate for an application is vital for both the image providers and users. A straightforward method for doing this is to assess the coded images by human observation alone because they are the final perceivers of the images. Techniques employing this method are called subjective assessment. Subjective assessment is time-consuming, expensive, and controversial. As a result, the development of more accurate and robust objective image quality assessment methods is getting more attention nowadays (Rohaly, Libert, Corriveau & Webster, 2000).
There are three types of objective image assessment methods existing: full-reference (FR), reduced-reference (RR), and no-reference (NR), respectively. In the FR method, a reference or undistorted image is required to present for assessing the corresponding qualities of the distorted images. However, in most image processing applications, it is difficult, if not impossible, to gain an access to the original or undistorted images. RR method requires a partial presence of original image or some extracted features of original image to assess the quality. It is feasible to transmit the extracted RR feature data to the receiver through an ancillary channel of communication for this purpose (Wang, Sheikh & Bovik, 2003). However, in many practical applications, the reference image is not available and the use of an ancillary channel could also be too expensive. Thus, the NR quality assessment approaches are more desired in most situations.
In the last three decades, a great deal of effort has been made to develop objective image and video quality assessment methods, especially on FR metrics. The most commonly used FR objective image and video distortion (or quality) metrics are mean squared error (MSE) and peak signal-to-noise ratio (PSNR). MSE and PSNR are widely used because they are simple to calculate, have clear physical meanings, and are mathematically easy to deal with for optimization purpose. However, the metrics have been widely criticized as well for not correlating well with human perceived quality measurement (Wang, Bovik, Sheikh, & Simoncelli, 2004), though the phase I test of the video quality experts group (VQEG) (Rohaly, Libert, Corriveau & Webster, 2000) achieved some success. Therefore the VQEG is continuing the work on Phase II for testing RR and NR assessment methods on television and Internet multimedia signals.