王克強(qiáng) 張雨帥 王保群

摘 要：多曝光圖像融合技術(shù)是將一組場(chǎng)景相同但曝光程度不同的圖像序列直接融合成為一幅含有更多場(chǎng)景細(xì)節(jié)信息的高質(zhì)量圖像。針對(duì)現(xiàn)有算法局部對(duì)比度差和色彩失真的問題，結(jié)合Retinex理論模型提出了一種新的多曝光圖像融合算法。首先，基于Retinex理論模型，利用光照估計(jì)算法將曝光序列圖像分為入射光分量序列和反射光分量序列，然后分別采用不同的融合方法對(duì)這兩組序列進(jìn)行處理。對(duì)于入射光分量，要保證場(chǎng)景的全局亮度的變化特性并且削弱過曝光和欠曝光區(qū)域的影響;而對(duì)于反射光分量，要采用適度曝光的評(píng)價(jià)參數(shù)來(lái)更好地保留場(chǎng)景的色彩及細(xì)節(jié)信息。分別從主觀和客觀兩方面對(duì)所提算法進(jìn)行了分析。實(shí)驗(yàn)結(jié)果表明，同傳統(tǒng)基于圖像域合成的算法相比，該算法在結(jié)構(gòu)相似度（SSIM）上平均提升了1.7%，另外在圖像色彩和局部細(xì)節(jié)上的處理效果更好。

關(guān)鍵詞：高動(dòng)態(tài)范圍成像;多曝光圖像;圖像融合;Retinex;曝光適度

Abstract： Multi-exposure image fusion technology directly combines a sequence of images with the same scene but different exposure levels into a high-quality image with more details of scene. Aiming at the problems of poor local contrast difference and color distortion of existing algorithms， a new multi-exposure image fusion algorithm was proposed based on Retinex theoretical model. Firstly， based on Retinex theoretical model， the exposure sequence images were divided into an illumination component sequence and a reflection component sequence by using the illumination estimation algorithm， and then two sets of sequences were processed by different fusion methods. For the illumination component， the variation characteristics of global brightness of scene were guaranteed and the effects of overexposed and underexposed regions were weakened， while for the reflection component， the evaluation parameters of moderate exposure were used to better preserve the color and detail information of scene. The proposed algorithm was analyzed from both subjective and objective aspects. The experimental results show that compared with traditional algorithm based on image domain synthesis， the proposed algorithm has an average increase of 1.7% in Structural SIMilarity （SSIM） and has better effect in the processing of image color and local details.

Key words： high dynamic range imaging; multi-exposure image; image fusion; Retinex; well-exposedness

0 引言

普通數(shù)碼相機(jī)成像的動(dòng)態(tài)范圍遠(yuǎn)低于現(xiàn)實(shí)場(chǎng)景的動(dòng)態(tài)范圍，其捕捉的畫面很難完整地呈現(xiàn)現(xiàn)實(shí)場(chǎng)景的所有細(xì)節(jié)信息。將場(chǎng)景多個(gè)不同曝光程度的低動(dòng)態(tài)范圍（Low Dynamic Range， LDR）圖像融合成高動(dòng)態(tài)范圍（High Dynamic Range，HDR）圖像是克服相機(jī)有限的動(dòng)態(tài)范圍并降低照片中噪聲的有效方法，這種成像技術(shù)稱為HDR成像[1]。由于拍攝時(shí)的相機(jī)抖動(dòng)以及場(chǎng)景內(nèi)可能存在運(yùn)動(dòng)對(duì)象，需要對(duì)所有LDR圖像先進(jìn)行對(duì)齊[2]，然后根據(jù)預(yù)定義的參考圖像同步所有運(yùn)動(dòng)對(duì)象[3]，再將校正后的圖像合成HDR圖像，以包括所有LDR圖像的細(xì)節(jié)，最后使用色調(diào)映射算法[4]將HDR圖像最終轉(zhuǎn)換為L(zhǎng)DR圖像，以便通過常規(guī)顯示設(shè)備來(lái)展示。

除了HDR成像技術(shù)之外，目前更為流行的一種技術(shù)是多曝光圖像融合。不同于HDR成像那樣需要生成中間HDR圖像，多曝光圖像融合技術(shù)直接從所有LDR圖像中合成信息量更大且視覺效果更好的LDR圖像。Mertens等[5]在多尺度圖像分解下，利用曝光程度、對(duì)比度和飽和度的三個(gè)質(zhì)量評(píng)價(jià)參數(shù)來(lái)確定給定像素對(duì)最終合成圖像的貢獻(xiàn)程度，利用多分辨率融合有效地保留了全局對(duì)比度，但是局部對(duì)比度較低。Zhang等[6]提出了一種基于梯度信息的曝光融合方案，認(rèn)為當(dāng)像素獲得更好的曝光狀態(tài)時(shí)，梯度幅度變得更大，并且隨著像素接近曝光過度/曝光不足而逐漸減小。Shen等[7]提出了一種基于概率模型的多曝光圖像融合算法，建立了一種廣義隨機(jī)游動(dòng)框架，通過將融合問題表示為概率估計(jì)，計(jì)算出兩種質(zhì)量度量下的全局最優(yōu)解，以此得到最終的合成圖像。Ma等[8]提出的方法將每個(gè)彩色圖像分解為三個(gè)概念上獨(dú)立的分量：信號(hào)強(qiáng)度、信號(hào)結(jié)構(gòu)和平均強(qiáng)度，對(duì)這三個(gè)分量分別進(jìn)行處理后得到融合圖像。