【摘要】：基于最小能量誤差得到左右視圖的視差圖,分別對(duì)左、右視圖和視差圖進(jìn)行4尺度、12個(gè)方向的可控金字塔分解,每一幅圖像可得到1條高頻子帶和48條方向子帶。對(duì)左、右視圖分解后相對(duì)應(yīng)的48對(duì)方向子帶進(jìn)行二元廣義高斯分布擬合,提取其形狀參數(shù)和尺度參數(shù),并提取所有方向子帶的跨尺度相關(guān)性、空間相關(guān)性等特征信息,將這些特征輸入支持向量回歸(SVR)訓(xùn)練預(yù)測(cè)得到立體圖像質(zhì)量評(píng)分。結(jié)果表明該質(zhì)量評(píng)價(jià)模型在LIVE 3D數(shù)據(jù)庫上的性能指標(biāo)斯皮爾曼等級(jí)相關(guān)系數(shù)(SROCC)、線性皮爾遜相關(guān)系數(shù)(CC)均在0.93以上,與人類的主觀評(píng)價(jià)具有較好的一致性。
[Abstract]:Based on the minimum energy error, the parallax graphs of left and right views are obtained, and the left, right and parallax images are decomposed with 4 scales and 12 directions respectively. One high frequency subband and 48 directional subbands can be obtained for each image. After decomposing the left and right view, 48 opposite direction sub-bands are fitted with binary generalized Gao Si distribution, and the shape parameters and scale parameters are extracted, and the cross-scale correlation, spatial correlation and other characteristic information of all directional sub-bands are extracted. These features are input into support vector regression (SVR) training to predict stereo image quality. The results show that the Spelman grade correlation coefficient (SROCC), linear Pearson correlation coefficient (CC) of the quality evaluation model in LIVE 3D database is above 0.93, which is in good agreement with human subjective evaluation.
【作者單位】： 江南大學(xué)物聯(lián)網(wǎng)工程學(xué)院輕工過程先進(jìn)控制教育部重點(diǎn)實(shí)驗(yàn)室;安徽省明光市第三中學(xué);
【基金】：國(guó)家自然科學(xué)基金,61170120~~
【分類號(hào)】：TP391.41

【相似文獻(xiàn)】

相關(guān)期刊論文 前10條

1 韓偉;;日本立體圖像技術(shù)近十年的回顧與前瞻[J];有線電視技術(shù);2011年07期

2 楊s，

本文編號(hào)：2251814