Blind Image Deconvolution Using Specified 2-D HPF for Feature Extraction and Conjugate Gradient Method in Frequency Domain

Takanori FUJISAWA  Masaaki IKEHARA  

IEICE TRANSACTIONS on Fundamentals of Electronics, Communications and Computer Sciences   Vol.E100-A   No.3   pp.846-853
Publication Date: 2017/03/01
Online ISSN: 1745-1337
DOI: 10.1587/transfun.E100.A.846
Type of Manuscript: PAPER
Category: Image
deblurring,  feature extraction,  optimization,  

Full Text: PDF>>
Buy this Article

Image deconvolution is the task to recover the image information that was lost by taking photos with blur. Especially, to perform image deconvolution without prior information about blur kernel, is called blind image deconvolution. This framework is seriously ill-posed and an additional operation is required such as extracting image features. Many blind deconvolution frameworks separate the problem into kernel estimation problem and deconvolution problem. In order to solve the kernel estimation problem, previous frameworks extract the image's salient features by preprocessing, such as edge extraction. The disadvantage of these frameworks is that the quality of the estimated kernel is influenced by the region with no salient edges. Moreover, the optimization in the previous frameworks requires iterative calculation of convolution, which takes a heavy computational cost. In this paper, we present a blind image deconvolution framework using a specified high-pass filter (HPF) for feature extraction to estimate a blur kernel. The HPF-based feature extraction properly weights the image's regions for the optimization problem. Therefore, our kernel estimation problem can estimate the kernel in the region with no salient edges. In addition, our approach accelerates both kernel estimation and deconvolution processes by utilizing a conjugate gradient method in a frequency domain. This method eliminates costly convolution operations from these processes and reduces the execution time. Evaluation for 20 test images shows our framework not only improves the quality of recovered images but also performs faster than conventional frameworks.