Patch-Based Video Denoising With Optical Flow Estimation

ABSTRACT:

A novel image sequence denoising algorithm is presented. The proposed approach takes advantage of the self-similarity and redundancy of adjacent frames. The algorithm is inspired by fusion algorithms, and as the number of frames increases, it tends to a pure temporal average. The use of motion compensation by regularized optical flow methods permits robust patch comparison in a spatiotemporal volume. The use of principal component analysis ensures the correct preservation of fine texture and details. An extensive comparison with the state-of-the-art methods illustrates the superior performance of the proposed approach, with improved texture and detail reconstruction.

PROJECT OUTPUT VIDEO:

EXISTING SYSTEM:

• In the literature, image fusion is not directly of interest in the removal of noise but in a more general restoration of the image, that is, deblurring, increase of detail or even of resolution. The key of these approaches is the use of a global registration, more robust to noise, blur and color or compression artifacts and, additionally, providing subpixel accuracy.

• These global registration techniques usually rely on feature matching, for example SIFT, and on a parametric registration, either using an affinity or an homography. The viewfinder alignment performs such a registration by an affine function, with the important characteristic of being extremely fast.

DISADVANTAGES OF EXISTING SYSTEM:

• The general approach is the use of an homography. It must be noted that an homography is valid only for planar scenes or if the optical center is not modified.

PROPOSED SYSTEM:

• We propose a new algorithm making use of motion estimation algorithms and patch based methods for denoising. Our method is inspired by image fusion algorithms in the sense that it tends to a fusion algorithm as the temporal sampling of the sequence gets dense and the motion estimation or global registration is able to perfectly register the frames and no occlusions are present.

• As this is an ideal scenario, our algorithm compensates the failure of these requirements by introducing patch comparison and denoising in an adapted PCA based transform. Unlike VBM4D the motion estimation used by our algorithm relies on the optical flow constraint (OFC), that is, we suppose that the color of each pixel remains constant along its trajectory through the sequence.

• The optical flow is used to warp adjacent frames and not only for compensating neighborhoods. Thus, the subpixel accuracy improves the patch comparison and averaging. Results in motion estimation are far from being totally satisfactory, there are many unsolved issues as occlusions, non translational motions, non color constancy, etc. Despite these limitations, we will show that OFC algorithms are a useful tool for denoising.

ADVANTAGES OF PROPOSED SYSTEM:

• An extensive comparison with the state-of-the-art methods illustrates the superior performance of the proposed approach, with improved texture and detail reconstruction.

MODULES:

• Video Acquisition

• Motion Compensation

• Denoising

• Performance Analysis

MODULE DESCRIPTION:

Image Acquisition:

In this module, first we develop the Image Acquisition module. The initial module does the process of Pre-processing steps. The Input Video are acquired from Gallery. Then video is converted into frames for further process. These output frames are used as the input for the next modules, for the evaluation of our proposed model.

Motion Compensation:

In this stage, motion compensation is done by motion estimation. First, the optical flow between I _k and adjacent frames in a temporal neighborhood is computed and used for warping these frames onto I _k. If registration was accurate and the sequence free of occlusions, a temporal average in this aligned data would be optimal, even if the noise reduction would slowly decrease as 1/M, where M is the number of adjacent frames involved in the process. Generally, this will not be the case, inaccuracies and errors in the computed flow and the presence of occlusions make this temporal average likely to blur the sequence and have artifacts near occlusions. The proposed approach tends to solve these limitations. Occlusions are detected depending on the divergence of the computed flow: negative divergence values indicate occlusions. Additionally, the color difference is checked after flow compensation. A large difference indicates occlusion, or at least failure of the color constancy assumption. We combine both criteria for a pixel x=(x, y) and the computed flow between I0 and I1. These occluded points having a negative divergence of the flow and a large color difference after flow compensation are located near the discontinuities of the motion field. In this patch wise motion compensated is performed.

Denoising:

In this stage, Denoising is performed. After extract the motion compensation between noised i and i+1frame, PCA is applied in motion compensated frames. Use of PCA for patch denoising preserves fine and texture details.

Performance Analysis:

In this module, we show the graph results for the Performance analysis module. We show the performance analysis results for 1) PSNR for Noise Frames 2) PSNR for Motion Estimated Frames and 3) PSNR for Denoised Frames. Peak signal-to-noise ratio, often abbreviated PSNR, is an engineering term for the ratio between the maximum possible power of a signal and the power of corrupting noise that affects the fidelity of its representation. Because many signals have a very wide dynamic range, PSNR is usually expressed in terms of the logarithmic decibel scale. PSNR is most commonly used to measure the quality of reconstruction of lossy compression codecs (e.g., for image compression). The signal in this case is the original data, and the noise is the error introduced by compression. When comparing compression codecs, PSNR is an approximation to human perception of reconstruction quality. Although a higher PSNR generally indicates that the reconstruction is of higher quality, in some cases it may not. One has to be extremely careful with the range of validity of this metric; it is only conclusively valid when it is used to compare results from the same codec (or codec type) and same content

SYSTEM REQUIREMENTS:

HARDWARE REQUIREMENTS:

• System          : Pentium IV 2.4 GHz.

• Hard Disk     : 40 GB.

• Floppy Drive : 1.44 Mb.

• Monitor          : 15 VGA Colour.

• Mouse            : Logitech.

• Ram                : 512 Mb.

SOFTWARE REQUIREMENTS:

• Operating system : Windows XP/7.

• Coding Language : MATLAB

• Tool                       : MATLAB R2013A

REFERENCE:

Antoni Buades, Jose-Luis Lisani, and Marko Miladinovc, “Patch-Based Video Denoising With Optical Flow Estimation”, IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 25, NO. 6, JUNE 2016.