International Journal of Emerging Technology and Advanced Engineering Website: www.ijetae.com (ISSN 2250-2459, ISO 9001:2008 Certified Journal, Volume 4, Issue 7, July 2014) 822 Image Quality Assessment Using Non-Linear Multi- Metric Fusion Approach Raman Gupta 1 , Dipti Bansal 2 , Charanjit Singh 3 1 M. Tech. Student, Punjabi University, Patiala, India 2,3 Assistant Professor, Punjabi University, Patiala, India Abstract— This paper presents a new approach for objective image quality assessment using multi-metric fusion. The current research is motivated by the observation that there is no single metric that provides the best performance scores in all conditions. To attain MMF, we approve an analytical formula based approach. First, we collect a large number of test images, each of which has a score labeled by human observers and scores linked with different metrics. The new MMF score is evaluated as the nonlinear combination of multiple metrics. It is shown by experimental results that the suggested MMF metric outperforms all existing metrics by a significant margin. Keywords—Image Quality Assessment (IQA), Multi-metric fusion (MMF), Mean Opinion Score (MOS). I. INTRODUCTION Visual image quality assessment method play important role in numerous image processing applications. There are mainly two types of image quality measures. First is Subjective image quality measure in which Mean Opinion Score (MOS) is estimated with the help of human observers. Second is Objective image quality measure in which mathematical expression is provided to estimate the quality of an image. However, later technique is proved to be better than previous because of two reasons. First, they are easy to assess due to its low computational complexity [1]. Second, they are independent of human visual perception. Although, viewing conditions plays important role in human perception of visual image quality, subjective methods are inconvenient and costlier to use due to involvement of number of observers and their numbers of corresponding results. Mean-square error (MSE) and Peak signal-to-noise ratio (PSNR) are two extensively used objective quality measures among objective visual assessment approaches. But, they may not correlate with human perception very well [2]. During the last decades, various new quality methods has been suggested as a replacements. Examples are SSIM [3], MSSIM [4], UQI [5], PSNR [6], PSNRHVS [7], IFC [8], FSIM [9]. But none of them significantly outperforms other. Few metrics may be superior for one image distortion type but inferior for other. Thus the idea to fuse a flexible number of metrics is arisen. To attain multi-metric fusion, we approve analytic formula based approach. First, we collect a large number of test images. Each image has a score labeled by human observers. Second, we evaluate scores associated with different metrics for each image. Then, we calculate the new MMF score as a non-linear combination of multiple metrics II. PREVIOUS WORK Luo and Huitao [10] proposed a two-step algorithm to measure image quality. First, a face detection algorithm is used to sense certain object types from the image. Second, the spectrum distribution of the sensed object is compared with a trained model to conclude its quality score. The limitation is that it predominantly applies to objects that contain human faces. Suresh et al. [11]. Proposed the practice of a machine learning method to evaluate the visual quality of JPEG-coded images. Features are extracted by bearing in mind human visual sensitivity factors, such as edge length, edge amplitude and background luminance. The visual quality of an image is then figured using the class number and their projected posterior probability. Experimental results shown that his approach performed better than other metrics, but it is only applicable to JPEG-coded images. The machine learning tool was also used in objective image quality metric. For example, Narwaria and Lin [12] proposed to use singular vectors out of singular value decomposition (SVD) as structures to quantify the major structural information in images. Then, they applied support vector regression (SVR) for image quality prediction, where the SVR method has the ability to study complex data patterns and maps complex features into a proper score. Tsung-Jung Liu et.al. [2] has proposed a multi-metric fusion approach for objective image quality assessment.