本文選題：圖像盲去運(yùn)動(dòng)模糊 + 二階梯度��； 參考：《華東師范大學(xué)》2017年碩士論文

【摘要】：圖像是人類(lèi)傳遞信息的重要媒介,在人們的日常生產(chǎn)生活中廣泛應(yīng)用。然而,在圖像拍攝過(guò)程中容易發(fā)生成像設(shè)備與目標(biāo)對(duì)象間的相對(duì)運(yùn)動(dòng),造成拍攝的圖像模糊不清,嚴(yán)重制約圖像的使用。圖像去運(yùn)動(dòng)模糊的目的是用一定數(shù)值化方法,從退化的模糊圖像中恢復(fù)出清晰的圖像。圖像去運(yùn)動(dòng)模糊是圖像處理領(lǐng)域的重要分支,特別是近年來(lái)人工智能的興起,對(duì)其進(jìn)行研究有極其重要的科研價(jià)值和現(xiàn)實(shí)意義。圖像去運(yùn)動(dòng)模糊問(wèn)題可以分為兩種類(lèi)型:事先可以確定模糊核的非盲去運(yùn)動(dòng)模糊和事先不能確定模糊核的盲去運(yùn)動(dòng)模糊。模糊核未知的圖像盲去運(yùn)動(dòng)模糊問(wèn)題,由于其未知量遠(yuǎn)大于已知量,具有嚴(yán)重的不適定性,求解極其困難。但是大多數(shù)的運(yùn)動(dòng)模糊圖像模糊核都是未知的,因此對(duì)圖像盲去運(yùn)動(dòng)模糊問(wèn)題進(jìn)行研究很有必要。本文聚焦于單幅圖像的盲去運(yùn)動(dòng)模糊問(wèn)題,主要工作包括以下幾個(gè)方面:(1)通過(guò)對(duì)圖像運(yùn)動(dòng)模糊成因的總結(jié),基于對(duì)自然圖像統(tǒng)計(jì)特征的分析,本文首次提出將圖像二階梯度特征應(yīng)用于圖像盲去運(yùn)動(dòng)模糊領(lǐng)域。在前人研究的基礎(chǔ)上,本文提出一種基于圖像二階梯度歸一化l1范數(shù)稀疏先驗(yàn)的自然圖像盲去運(yùn)動(dòng)模糊方法,并且給出一種有效的快速求解算法。(2)由于特殊圖像,如文本圖像,并不符合一般自然圖像的統(tǒng)計(jì)特征,往往需要單獨(dú)設(shè)計(jì)相應(yīng)的去運(yùn)動(dòng)模糊方法。通過(guò)對(duì)自然圖像和特殊圖像特征的分析,本文又提出一種對(duì)自然圖像和特殊圖像都有效的,基于圖像二階梯度和暗通道稀疏先驗(yàn)的圖像盲去運(yùn)動(dòng)模糊方法。我們用一系列的實(shí)驗(yàn)驗(yàn)證本文方法的有效性,并與其它現(xiàn)有方法做比較。實(shí)驗(yàn)結(jié)果表明,本文的方法能夠有效去除圖像運(yùn)動(dòng)模糊,提升圖像的復(fù)原質(zhì)量。本文的工作對(duì)自然圖像和特殊圖像的盲去運(yùn)動(dòng)模糊問(wèn)題做出了相應(yīng)的改進(jìn)。
[Abstract]:Image is an important medium for human to transmit information, which is widely used in people's daily production and life. However, the relative motion between the imaging equipment and the target object is easy to occur in the process of image capturing, which results in the blurred image, which seriously restricts the use of the image. The purpose of image demotion blur is to restore the clear image from the degraded blurred image by a certain numerical method. Image demotion blur is an important branch in the field of image processing, especially the rise of artificial intelligence in recent years, the research on it has extremely important scientific research value and practical significance. The problem of image demotion blur can be divided into two types: the non-blind deblurring of the blur kernel can be determined in advance and the blind deblurring of the blur kernel can not be determined in advance. Because the unknown unknowns of image demotion blur problem is much larger than the known one, it is very difficult to solve the problem because of its serious ill-posed nature. However, most of the motion-blurred image blur cores are unknown, so it is necessary to study the image blind motion blur problem. This paper focuses on the blind motion blur of a single image. The main work includes the following aspects: 1) based on the analysis of the statistical features of natural images, the causes of motion blur are summarized. In this paper, the second order gradient feature of image is applied to the field of blind motion blur for the first time. On the basis of previous studies, this paper proposes a blind deblurring method for natural images based on image second-order gradient normalization L _ 1-norm sparse priori, and gives an efficient and fast algorithm to solve motion blur for special images, such as text images. Because it does not accord with the statistical characteristics of general natural images, it is often necessary to design the corresponding de-motion blur method separately. By analyzing the features of natural and special images, this paper proposes a blind motion blur method based on image second-order gradient and dark channel sparse priori, which is effective for both natural and special images. We use a series of experiments to verify the effectiveness of this method and compare it with other existing methods. Experimental results show that the proposed method can effectively remove image motion blur and improve image restoration quality. In this paper, the blind deblurring problem of natural and special images is improved.
【學(xué)位授予單位】：華東師范大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2017
【分類(lèi)號(hào)】：TP391.41

【參考文獻(xiàn)】

相關(guān)期刊論文 前1條

1 許鋒,盧建剛,孫優(yōu)賢;神經(jīng)網(wǎng)絡(luò)在圖像處理中的應(yīng)用[J];信息與控制;2003年04期

，

本文編號(hào)：1855967