Turbulence makes the image suffer from geometric distortion, pixel deviation and blur. This paper focuses on image restoration under atmospheric turbulence. To improve the image quality, we revisit the problem by a two-phase method. According to the distortion model analysis, we first combine affine transformation with non-rigid registration to suppress global motion and local pixel deviation. To improve the registration speed, the cost function is optimized by L-BFGS algorithm. Next, a multi-frame blind deconvolution algorithm is employed to restore the registered frames, and get a final output. The experimental results clearly demonstrate the effectiveness of the proposed method. It can effectively alleviate blur and distortions, improve visual quality and recovery speed significantly.
The goal of image restoration is to reconstruct the original scene from a degraded observation. It is a critical and
challenging task in image processing. Classical restorations require explicit knowledge of the point spread function and a
description of the noise as priors. However, it is not practical for many real image processing. The recovery processing
needs to be a blind image restoration scenario. Since blind deconvolution is an ill-posed problem, many blind restoration
methods need to make additional assumptions to construct restrictions. Due to the differences of PSF and noise energy,
blurring images can be quite different. It is difficult to achieve a good balance between proper assumption and high
restoration quality in blind deconvolution. Recently, machine learning techniques have been applied to blind image
restoration. The least square support vector regression (LSSVR) has been proven to offer strong potential in estimating
and forecasting issues. Therefore, this paper proposes a LSSVR-based image restoration method. However, selecting the
optimal parameters for support vector machine is essential to the training result. As a novel meta-heuristic algorithm, the
fruit fly optimization algorithm (FOA) can be used to handle optimization problems, and has the advantages of fast
convergence to the global optimal solution. In the proposed method, the training samples are created from a
neighborhood in the degraded image to the central pixel in the original image. The mapping between the degraded image
and the original image is learned by training LSSVR. The two parameters of LSSVR are optimized though FOA. The
fitness function of FOA is calculated by the restoration error function. With the acquired mapping, the degraded image
can be recovered. Experimental results show the proposed method can obtain satisfactory restoration effect. Compared
with BP neural network regression, SVR method and Lucy-Richardson algorithm, it speeds up the restoration rate and
performs better. Both objective and subjective restoration performances are studied in the comparison experiments.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.