光子学报, 2018, 47 (12): 1210002, 网络出版: 2019-01-10
基于立体图像感兴趣区域及对比度的舒适度评价模型
Comfort Evaluation Model Based on Region of Interest and Contrast of Stereo Images
双目立体图像 视觉舒适度 对比度 感兴趣区域 客观评价 颜色空间 最小二乘法 Binocular stereo images Degree of visual comfortable Contrast Region of interest Objective evaluation Color space Least square method
摘要
针对立体图像舒适度难以有效地进行客观评价的问题, 结合人眼视觉注意机制, 提出了基于区域对比度的舒适度评价模型.根据显著图与视差图提取感兴趣区域作为前景区域; 量化前景和后景区域颜色空间, 估计空间加权区域对比度, 计算前景区域视差角、宽度角; 根据主观评价值利用最小二乘法拟合曲线得出客观评价模型.对比视差+宽度模型可知, 模型预测值与主观评价值的Pearson相关系数、Kendall相关系数较原模型分别提高了8.1%、3.9%, 且平均绝对值误差减小了13%, 均方根误差减小了22.1%.本文模型的普适性更优, 结果更接近主观评价值.
Abstract
A comfort evaluation model was proposed based on regional contrast, aiming at the problem that the stereo image comfort is difficult to effectively evaluate objectively, combined with the human visual attention mechanism. The region of interest was extracted as the foreground region according to the saliency map and the disparity map. Then, the color spaces of the foreground and background region were quantized, and the regional contrast with the spatial weighted was estimated. The parallax angle and width angle of the foreground area were calculated. Finally, according to the subjective evaluation values, the objective evaluation model was obtained by fitting the curve with least-square method. Comparing the D+W model, the Pearson correlation coefficient and the Kendall correlation coefficient of the model prediction value and the subjective evaluation value are 8.1% and 3.9% higher respectively than the original model. The average absolute value error is reduced by 13%, and the root mean square error is reduced by 22.1%. The experimental results show that the model has better universality and the result is closer to the subjective evaluation value.
权巍, 赵云秀, 韩成, 丁莹, 姜珊, 李波. 基于立体图像感兴趣区域及对比度的舒适度评价模型[J]. 光子学报, 2018, 47(12): 1210002. QUAN Wei, ZHAO Yun-xiu, HAN Cheng, DING Ying, JIANG Shan, LI Bo. Comfort Evaluation Model Based on Region of Interest and Contrast of Stereo Images[J]. ACTA PHOTONICA SINICA, 2018, 47(12): 1210002.