DSP-based image real-time dehazing optimization for improved dark-channel prior algorithm
- PDF / 1,241,036 Bytes
- 10 Pages / 595.276 x 790.866 pts Page_size
- 55 Downloads / 219 Views
ORIGINAL RESEARCH PAPER
DSP‑based image real‑time dehazing optimization for improved dark‑channel prior algorithm Jinzheng Lu1 · Chuan Dong1 Received: 23 February 2019 / Accepted: 27 November 2019 © Springer-Verlag GmbH Germany, part of Springer Nature 2019
Abstract To solve the problem of non-real-time processing of image dehazing using traditional dark-channel prior algorithm, this work studies image real-time penetrating fog optimization technologies based on digital signal processor (DSP) devices. Using jointed optimization mechanism between algorithm and device, we can achieve real-time processing. During algorithm optimization, mean filter characterized low computation substitutes the guided filter which is the most complex in dark-channel algorithm for dehazing. In optimization of image processing task under the embedded device, we empirically construct two-step optimization strategy for raising speed of processing. Thereupon, the awful division calculation for DSP device is achieved approximately by multiplication after the reciprocal operation. We utilize the specified template which is considerably designed to realize mean filter. Thus, the division factor in the template can be calculated innovatively via shift instructions featured on DSP. The experimental results show that the optimization solution provided has realized real-time image dehazing processing for standard-definition and high-definition at frame rate of 25 fps over C6748 pure DSP device featured 456 MHz clock, at the same time the effect of penetrating fog is not remarkably degraded. The optimization methods or ideas can easily be transplanted to similar platform. Keywords Image dehazing · Dark-channel prior · Software pipeline · Single instruction multiple data (SIMD) · Intrinsic instructions
1 Introduction Because of the influence of fog, haze, smoke and other factors, the imaging results of outdoor scenes are usually blurred. The degraded image’s contrast is reduced and its color is dim. Therefore, the missed key information has a significant impact on image intuitive experience, target detection and tracking, target recognition and so on [1, 2]. The technology of image sharpening has imperious demands in the fields of video surveillance, computer vision and so on. Image dehazing can improve the utilization value of foggy images and increase the image perception experience [3]. Image penetrating fog can significantly improve scene intelligibility, correct color deviation, provide accurate target information for machine vision, and improve the algorithm * Jinzheng Lu [email protected] 1
School of Information Engineering, Southwest University of Science and Technology, 59 Qinglong Road, Mianyang 621010, Sichuan, People’s Republic of China
performance of feature extraction, filtering or component analysis. There are many methods for image dehazing, including image enhancement, image fusion and image restoration [4]. The effect of first two techniques is limited in practice application, or the real-time image processing is common
Data Loading...
