CS Events Monthly View

PhD Defense

Single Image deblurring with or without face prior and its applications


Download as iCal file

Friday, February 13, 2015, 04:30pm


The motion blur is one of the most difficult challenges in photography, which is generated from the relative motion between the sensor and the scene during the exposure time. These blur artifacts degrade the visual experience, and the performance of various application, such as, object detection, facial analysis. Therefore, it is desirable to remove the blur and restore sharp and clean images computationaly. Our work focuses on the general single image deblurring, and face image deblurring with face prior.

State-of-the-art single image deblurring techniques are sensitive to image noise. Even a small amount of noise, which is inevitable in low-light conditions, can degrade the quality of blur kernel estimation dramatically. The recent approach of Tai and Lin~cite{Tai12} tries to iteratively denoise and deblur a blurry and noisy image. However, as we show in this work, directly applying image denoising methods often partially damages the blur information that is extracted from the input image, leading to biased kernel estimation. We propose a new method for handling noise in blind image deconvolution based on new theoretical and practical insights. Our key observation is that applying a directional low-pass filter to the input image greatly reduces the noise level, while preserving the blur information in the orthogonal direction to the filter. Based on this observation, our method applies a series of directional filters at different orientations to the input image, and estimates an accurate Radon transform of the blur kernel from each filtered image. Finally, we reconstruct the blur kernel using inverse Radon transform. Experimental results on synthetic and real data show that our algorithm achieves higher quality results than previous approaches on blurry and noisy images.

The human face is one of the most essential focuses in numerous applications. Although significant progress has been made in the image deblurring area, few of them can obtain promising results on blurry face images. Many state-of-the-art single image deblurring approaches estimate the blur kernel based on analyzing the edge profiles of the input image. They first identify informative and less ambiguous edges, and use their observed blurry and estimated sharp profiles in various optimization formulations to estimate the blur kernel. Thus, selecting good edges is essential for these approaches to achieve accurate blur kernel estimation. However, the detection of strong edges is very difficult on human faces, since the human faces do not contain as much texture as natural images. We proposed to utilize the global face structure information (i.e., face landmark localization) to help with the strong or salient edge detection. With more accurate edge detection, our method outperforms the state-of-the-art methods in extensive evaluations on various synthetic and real face image data. Landmark detection is the key step of our method, so we also evaluate the robustness of the landmark localization on the blurry images with various kernel sizes.

Facial expression is a significant application on face images. We illustrated the performances of facial expression recognition on blurry and the corresponding restored face images. To improve the general facial expression recognition performance, we also present a new idea to analyze facial expression by exploring the common and specific information among different expressions.

Speaker: Lin Zhong



Location : CBIM 22


Dimitris Metaxas (advisor), Ahmed Elgammal, Kostas Bekris and Dimitris Samaras (Stony Brook Univeristy)

Event Type: PhD Defense



Rutgers University