本文關(guān)鍵詞： 去隔行 運動檢測 降噪 并行計算 塊匹配　出處：《北京交通大學(xué)》2017年碩士論文　論文類型：學(xué)位論文

【摘要】：隨著多媒體技術(shù)的發(fā)展,數(shù)字電視系統(tǒng)和顯示設(shè)備對視頻信號的質(zhì)量要求越來越高,視頻后處理技術(shù)變得越來越重要。視頻信號的去隔行技術(shù)和視頻的降噪技術(shù)是視頻后處理技術(shù)不可或缺的兩種技術(shù)。去隔行技術(shù)是為了解決由于隔行掃描而產(chǎn)生的鋸齒效應(yīng)的問題,降噪算法在提高視頻信號質(zhì)量的同時,也成為其它視頻增強算法的預(yù)處理流程。因此,去隔行和降噪兩種算法的研究都有較強的實際意義。高計算復(fù)雜度的算法在現(xiàn)實中會遇到時間復(fù)雜度過高的瓶頸而無法得到推廣,因此研究視頻處理算法的并行設(shè)計擁有了較高的實際意義。本文針對去隔行技術(shù)提出了一種基于運動檢測的邊緣自適應(yīng)去隔行算法,針對三維塊匹配降噪算法的時間復(fù)雜度過高的情況,設(shè)計了針對三維塊匹配降噪算法的并行程序。本文的主要工作如下:(1)針對隔行掃描的視頻信號會產(chǎn)生鋸齒效應(yīng),本文在前人的基礎(chǔ)上提出了一種改進(jìn)的基于運動檢測的邊緣自適應(yīng)去隔行算法。該算法在運動檢測階段引入了前后五場信息作為運動檢測的信息源,更多的利用了相關(guān)場的運動信息,使運動檢測更加精確。在運動判定的閾值上,本文引入了一種動態(tài)閾值的方法,閾值的大小根據(jù)相對應(yīng)像素的灰度值動態(tài)變化,有效的降低了檢測誤差。(2)在得到去隔行運動檢測結(jié)果后,對場信息進(jìn)行插值得到視頻幀信息。本文在插值過程中引入了紋理檢測器,將運動信息劃分為紋理區(qū)域以及平滑區(qū)域,在兩個區(qū)域內(nèi)采用不同的插值方法,有效的保護了圖像邊緣信息和高頻分量。實驗表明,本文在運動檢測和插值過程中提出的改進(jìn)提高了去隔行的效果。(3)三維塊匹配降噪算法是降噪算法中最有效的算法之一,該算法無論在主觀上還是客觀上都有著很好的降噪效果,但是因為算法的計算復(fù)雜度較高,處理一幀低分辨率的視頻信息就需要數(shù)秒鐘的時間,制約了該算法在實際中的應(yīng)用�；诖�,本文針對三維塊匹配算法進(jìn)行了并行算法設(shè)計,重點研究和設(shè)計了塊匹配和硬閾值濾波模塊,實驗表明,本文設(shè)計的并行算法達(dá)到了較高的加速比。
[Abstract]:With the development of multimedia technology, digital TV system and display equipment of video signal quality requirements more and more high, video postprocessing technology becomes more and more important. The noise reduction technology of video signal and video deinterlacing technology are two kinds of techniques of video postprocessing technology indispensable. Deinterlacing technology is to solve the aliasing effect due to the interlaced problem, noise reduction algorithm can improve the video quality at the same time, also become the other video enhancement preprocessing procedure of the algorithm. Therefore, deinterlacing and noise reduction of the two algorithms research has a strong practical significance. The high computational complexity of the algorithm in real time complexity will encounter bottlenecks too high and can not be extended, so the research on video processing parallel design has high practical significance. This paper algorithm de interlacing technique is proposed based on motion detection The edge adaptive deinterlacing algorithm for complex 3D block matching algorithm for noise reduction in time, is designed for 3D block matching noise reduction algorithm for parallel program. The main work of this paper are as follows: (1) the interlaced video signal will produce sawtooth effect, on the basis of predecessors' proposed an improved edge detection based on motion adaptive deinterlacing algorithm. The algorithm introduces five field motion detection before and after the information as a source of information in the motion detection stage, more use of motion information of related field, make motion detection more accurate. In the movement to determine the threshold, this paper introduces a method of dynamic the threshold and threshold according to the size of the corresponding pixel value changes, effectively reduce the detection error. (2) in de interlacing motion detection results, the field information is inserted to video Frame information. This paper introduces a texture detector in the process of interpolation, motion information is divided into texture region and smooth region, different interpolation methods used in the two region, effectively protect the image edge information and the high frequency component. Experimental results show that the proposed improvements in motion detection and interpolation process to improve vibration the results. (3) three dimensional block matching algorithm is one of the most effective noise reduction algorithm of noise reduction algorithm, the algorithm in both subjective and objective has a good noise reduction effect, but because of the computational complexity of the algorithm is high, processing a frame low resolution video information requires the number of seconds, restrict the application of the algorithm in practice. Based on this, this paper 3D block matching algorithm of parallel algorithm design, the research and design of block matching and hard threshold filtering module, the test results show that the The parallel algorithm designed in this paper achieves a higher acceleration ratio.

【學(xué)位授予單位】：北京交通大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：TP391.41

【參考文獻(xiàn)】

相關(guān)期刊論文 前2條

1 郭靜;陳慶奎;;基于CUDA的快速圖像壓縮[J];計算機工程與設(shè)計;2010年14期

2 李軍;李艷輝;陳雙平;;CUDA架構(gòu)下的快速圖像去噪[J];計算機工程與應(yīng)用;2009年11期

相關(guān)碩士學(xué)位論文 前1條

1 年華;GPU通用計算與基于SIFT特征的圖像匹配并行算法研究[D];西安電子科技大學(xué);2010年

，

本文編號：1458784