2009 International Symposium on Intelligent Signal Processing and Communication Systems (ISPACS 2009) December 7-9, 2009 MP1-B-4 978-1-4244-5016-9/09/$25.00 c 2009 IEEE – 115 – Video Enhancement Using A Robust Iterative SRR Based On Andrew’s Sine Regularization Technique Vorapoj Patanavijit † Department of Computer and Network Engineering, Faculty of Engineering, Assumption University, Bangkok, Thailand Email: Patanavijit@yahoo.com Abstract- In this paper, we propose a alternative robust video enhancement algorithm using SRR based on the regularization ML technique. First, the classical registration process is used to estimate the relationship between the reference frame and other neighboring frames. Subsequently, the Andrew’s Sine norm is used for measuring the difference between the projected estimate of the high quality image and each low high quality image and for removing outliers in the data. Moreover, Tikhonov regularization is incorporated in the proposed framework in order to remove artifacts from the final answer and to improve the rate of convergence. Later, the reconstructed video frame is estimated by minimize the total cost function. Finally, experimental results are presented to demonstrate the outstanding performance of the proposed algorithm in comparison to several previously published methods using standard sequences such as Foreman and Susie that are corrupted by several noise models such as AWGN, Poisson Noise, Salt & Pepper Noise and Speckle Noise. I. INTRODUCTION In general, video enhancement from a single observation is limited as the aliasing in the low-resolution image cannot be effectively eliminated by the image interpolation. SRR (Super- Resolution Reconstruction) make use of subpixel shifts between frames in an image sequence and each of them provides a unique view of the scene. Consequently, it can efficiently fuse the useful information and suppress the aliasing to generate a better high resolution image. [1-4]. Reference [5] reviews several important results on regularized SRR algorithms available in the literature from the estimation point of view since the SRR estimation is one of the most importance parts of the SRR research areas and directly affects to the SRR performance. From the literature result, almost regularized SRR algorithms are based on the simple estimation techniques such as L1 Norm or L2 Norm Minimization. For normally distributed data, the L1 norm produces estimates with higher variance than the optimal L2 (quadratic) norm but the L2 norm is very sensitive to outliers because the influence function increases linearly and without bound. Consequence, Reference [5] proposed a robust iterative SRR algorithm based on Andrew’s Sine norm [6], designed to be more robust than L1 and L2 norm, for synthesis corrupted image such as Lena standard image. In this paper, we proposed the novel video enhancement framework using a robust iterative SRR based on an Andrew’s Sine stochastic estimation. Andrew’s Sine norm is used for measuring the difference between the projected estimate of the high quality image and each low high quality image and for removing outliers in the data. Moreover, Tikhonov regularization is incorporated in the proposed framework in order to remove artifacts from the final answer and to improve the rate of convergence. These experimental results demonstrate that our method’s performance is superior to what were proposed earlier [1-5]. The organization of this paper is as follows. Section 2 introduces the main concepts of estimation technique in SRR frameworks based on L1 and L2 norm minimization. Later, section 3 presents the proposed video enhancement algorithm using SRR based on Andrew’s Sine norm minimization with Tikhonov Regularization. Section 4 outlines the proposed solution and presents the comparative experimental results obtained by using the proposed Andrew’s Sine norm method and by using the L1 and L2 norm method. Finally, Section 5 provides the summary and conclusion. II. INTRODUCTION OF VIDEO ENHANCEMENT USING SRR ALGORITHM A block diagram of the proposed video enhancement framework based on SRR algorithm is illustrated in figure 1. The proposed SRR framework is the iterative process composed of 2 main parts: the registration parameter estimation and image estimation. First, the registration parameters, which used to wrap all low resolution images and used in the image estimation process are estimated from low resolution images (or observed sequences) and the registration parameters. This classical block based registration is used in this process. To reduce the computational complexity, each frame is separated into overlapping blocks and the registration Registration Estimation 1 LR Image 2 LR Image  LR Image N  SRR Estimated Process  Estimated SR Image Figure 1: A block diagram of the proposed video enhancement framework based on SRR algorithm † This work has been supported by Research Grant for New Scholar from TRF (Thai Research Fund) and CHE (Commission on Higher Education) under Assumption University (Thailand).