-
图像质量评价(IQA)方法需要考虑如何从主观视觉度量结果出发, 设计出符合该结果的客观图像质量评价方法, 应用到相关实际问题中. 本文从视觉感知特性出发, 量化色度和结构特征信息, 提出了基于色貌和梯度两个图像特征的图像质量客观评价模型. 两个色貌新指标(vividness和depth)是色度特征信息提取算子; 梯度算子用来提取结构特征信息. 其中, vividness相似图一方面作为特征提取算子计算失真图像局部质量分数, 另一方面作为图像全局权重系数反应每个像素的重要程度. 为了量化所提模型的主要参数, 根据通用模型性能评价指标, 使用Taguchi实验设计方法进行优化. 为了验证该模型的性能, 使用4个常用图像质量数据库中的94幅参考图像和4830幅失真图像进行对比测试, 从预测精度、计算复杂度和泛化性进行分析. 结果表明, 所提模型的精度PLCC值在4给数据库中最低实现0.8455, 最高可以达到0.9640, 综合性能优于10个典型和近期发表的图像质量评估(IQA)模型. 研究结果表明, 所提模型是有效的、可行的, 是一个性能优异的IQA模型.With the rapid development of color image contents and imaging devices in various kinds of multimedia communication systems, conventional grayscale counterparts are replaced by chromatic ones. Under such a transition, the image quality assessment (IQA) model needs to be built by subjective visual measurement, designed in accordance with the results, and applied to the related practical problems. Based on the visual perception characteristics, chromaticity and the structure feature information are quantified, and an objective IQA model combining the color appearance and the gradient image features is proposed in this paper, namely color appearance and gradient similarity(CAGS) model. Two new color appearance indices, vividness and depth, are selected to build the chromatic similarity map. The structure information is characterized by gradient similarity map. Vividness map plays two roles in the proposed model. One is utilized as feature extractor to compute the local quality of distorted image, and the other is as a weight part to reflect the importance of local domain. To quantify the specific parameters of CAGS, Taguchi method is used and four main parameters, i.e., K V, K D, K Gand α, of this model are determined based on the statistical correlation indices. The optimal parameters of CAGS are K V= K D= 0.02, K G= 50, and α= 0.1. Furthermore, the CAGS is tested by utilizing 94 reference images and 4830 distorted images from the four open image databases (LIVE, CSIQ, TID2013 and IVC). Additionally, the influences of the 35 distortion types on IQA are analyzed. Massive experiments are performed on four publicly available benchmark databases between CAGS and other 10 state-of-the-art and recently published IQA models, for the accuracy, complexity and generalization performance of IQA. The experimental results show that the accuracy PLCC of the CAGS model can achieve 0.8455 at lowest and 0.9640 at most in the four databases, and the results about commonly evaluation criteria prove that the CAGS performs higher consistency with the subjective evaluations. Among the 35 distortion types, the two distortion types, namely contrast change and change of color saturation, CAGS and mostly IQA models have the worst influence on IQA, and the CAGS yields the highest top three rank number. Moreover, the SROCC values of CAGS for other distortion types are all larger than 0.6 and the number of SROCC value larger than 0.95 is 14 times. Besides, the CAGS maintains a moderate computational complexity. These results of test and comparison above show that the CAGS model is effective and feasible, and the corresponding model has an excellent performance.
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] -
数据库 原始图像数量 失真图像数量 失真类型 观察者 TID2013 25 3000 24 971 CSIQ 30 866 6 35 LIVE 29 779 5 161 IVC 10 185 4 15 
下载:
导出CSV
代号 参数表述 水准数 水准一 水准二 水准三 A KV 3 0.002 0.02 0.2 B KD 3 0.002 0.02 0.2 C KG 3 10 50 100 D α 3 0.1 0.5 1 下载:
导出CSV
实验序号 A B C D SROCC SROCC的S/N RMSE RMSE的S/N 1 1 1 1 1 0.9300 –0.6303 0.4113 7.7168 2 1 2 2 2 0.9192 –0.7318 0.4533 6.8723 3 1 3 3 3 0.9096 –0.8230 0.4825 6.3301 4 2 1 2 3 0.9171 –0.7517 0.4596 6.7524 5 2 2 3 1 0.9173 –0.7498 0.4672 6.6099 6 2 3 1 2 0.9291 –0.6388 0.4142 7.6558 7 3 1 3 2 0.9114 –0.8058 0.4735 6.4936 8 3 2 1 3 0.9279 –0.6500 0.4174 7.5890 9 3 3 2 1 0.9195 –0.7290 0.4481 6.9725 下载:
导出CSV
数据库 SSIM IW-SSIM IFC VIF MAD RFSIM FSIMC GSM CVSS MPCC Proposed TID2013 SROCC 0.7417 0.7779 0.5389 0.6769 0.7807 0.7744 0.8510 0.7946 0.8069 0.8452 0.8316 PLCC 0.7895 0.8319 0.5538 0.7720 0.8267 0.8333 0.8769 0.8464 0.8406 0.8616 0.8445 RMSE 0.7608 0.6880 1.0322 0.7880 0.6975 0.6852 0.5959 0.6603 0.6715 0.6293 0.6639 KROCC 0.5588 0.5977 0.3939 0.5147 0.6035 0.5951 0.6665 0.6255 0.6331 — 0.6469 CSIQ SROCC 0.8756 0.9213 0.7671 0.9195 0.9466 0.9295 0.9310 0.9108 0.9580 0.9569 0.9198 PLCC 0.8613 0.9144 0.8384 0.9277 0.9502 0.9179 0.9192 0.8964 0.9589 0.9586 0.9014 RMSE 0.1334 0.1063 0.1431 0.0980 0.0818 0.1042 0.1034 0.1164 0.0745 0.0747 0.1137 KROCC 0.6907 0.7529 0.5897 0.7537 0.7970 0.7645 0.7690 0.7374 0.8171 — 0.7487 LIVE SROCC 0.9479 0.9567 0.9259 0.9636 0.9669 0.9401 0.9599 0.9561 0.9672 0.9660 0.9734 PLCC 0.9449 0.9522 0.9268 0.9604 0.9675 0.9354 0.9503 0.9512 0.9651 0.9622 0.9640 RMSE 8.9455 8.3473 10.2643 7.6137 6.9073 9.6642 7.1997 8.4327 7.1573 7.4397 8.3251 KROCC 0.7963 0.8175 0.7579 0.8282 0.8421 0.7816 0.8366 0.8150 0.8406 — 0.8658 IVC SROCC 0.9018 0.9125 0.8993 0.8964 0.9146 0.8192 0.9293 0.8560 0.8836 — 0.9195 PLCC 0.9119 0.9231 0.9093 0.9028 0.9210 0.8361 0.9392 0.8662 0.8438 — 0.9298 RMSE 0.4999 0.4686 0.5069 0.5239 0.4746 0.6684 0.4183 0.6088 0.6538 — 0.4483 KROCC 0.7223 0.7339 0.7202 0.7158 0.7406 0.6452 0.7636 0.6609 0.6957 — 0.7488 权重平均 SROCC 0.8051 0.8376 0.6560 0.7750 0.8456 0.8306 0.8859 0.8438 0.8628 — 0.8737 PLCC 0.8321 0.8696 0.6786 0.8353 0.8752 0.8650 0.8987 0.8730 0.8820 — 0.8772 KROCC 0.6270 0.6662 0.5002 0.6158 0.6819 0.6575 0.7160 0.6775 0.7020 — 0.7044 直接平均 SROCC 0.8668 0.8921 0.7828 0.8641 0.9022 0.8658 0.9178 0.8794 0.9039 — 0.9111 PLCC 0.8769 0.9054 0.8071 0.8907 0.9164 0.8807 0.9214 0.8901 0.9021 — 0.9099 KROCC 0.6920 0.7255 0.6154 0.7031 0.7458 0.6966 0.7589 0.7097 0.7466 — 0.7526 下载:
导出CSV
数据库 失真类型 SSIM IW-SSIM IFC VIF MAD RFSIM FSIMC GSM CVSS MPCC Proposed TID2013 AGN 0.8671 0.8438 0.6612 0.8994 0.8843 0.8878 0.9101 0.9064 0.9401 0.8666 0.9359 ANC 0.7726 0.7515 0.5352 0.8299 0.8019 0.8476 0.8537 0.8175 0.8639 0.8187 0.8653 SCN 0.8515 0.8167 0.6601 0.8835 0.8911 0.8825 0.8900 0.9158 0.9077 0.7396 0.9276 MN 0.7767 0.8020 0.6932 0.8450 0.7380 0.8368 0.8094 0.7293 0.7715 0.7032 0.7526 HFN 0.8634 0.8553 0.7406 0.8972 0.8876 0.9145 0.9094 0.8869 0.9097 0.8957 0.9159 IN 0.7503 0.7281 0.6208 0.8537 0.2769 0.9062 0.8251 0.7965 0.7457 0.6747 0.8361 QN 0.8657 0.8468 0.6282 0.7854 0.8514 0.8968 0.8807 0.8841 0.8869 0.7931 0.8718 GB 0.9668 0.9701 0.8907 0.9650 0.9319 0.9698 0.9551 0.9689 0.9348 0.9218 0.9614 DEN 0.9254 0.9152 0.7779 0.8911 0.9252 0.9359 0.9330 0.9432 0.9427 0.9510 0.9466 JPEG 0.9200 0.9187 0.8357 0.9192 0.9217 0.9398 0.9339 0.9284 0.9521 0.8964 0.9585 JP2 K 0.9468 0.9506 0.9078 0.9516 0.9511 0.9518 0.9589 0.9602 0.9587 0.9160 0.9620 JPTE 0.8493 0.8388 0.7425 0.8409 0.8283 0.8312 0.8610 0.8512 0.8613 0.8571 0.8644 J2 TE 0.8828 0.8656 0.7769 0.8761 0.8788 0.9061 0.8919 0.9182 0.8851 0.8409 0.9250 NEPN 0.7821 0.8011 0.5737 0.7720 0.8315 0.7705 0.7937 0.8130 0.8201 0.7753 0.7833 Block 0.5720 0.3717 0.2414 0.5306 0.2812 0.0339 0.5532 0.6418 0.5152 0.5396 0.6015 MS 0.7752 0.7833 0.5522 0.6276 0.6450 0.5547 0.7487 0.7875 0.7150 0.7520 0.7441 CTC 0.3775 0.4593 0.1798 0.8386 0.1972 0.3989 0.4679 0.4857 0.2940 0.7814 0.4514 CCS 0.4141 0.4196 0.4029 0.3009 0.0575 0.0204 0.8359 0.3578 0.2614 0.7054 0.3711 MGN 0.7803 0.7728 0.6143 0.8486 0.8409 0.8464 0.8569 0.8348 0.8799 0.8766 0.8700 CN 0.8566 0.8762 0.8160 0.8946 0.9064 0.8917 0.9135 0.9124 0.9351 0.8174 0.9168 LCNI 0.9057 0.9037 0.8160 0.9204 0.9443 0.9010 0.9485 0.9563 0.9629 0.8095 0.9574 ICQD 0.8542 0.8401 0.6006 0.8414 0.8745 0.8959 0.8815 0.8973 0.9108 0.8596 0.9060 CHA 0.8775 0.8682 0.8210 0.8848 0.8310 0.8990 0.8925 0.8823 0.8523 0.8094 0.8768 SSR 0.9461 0.9474 0.8885 0.9353 0.9567 0.9326 0.9576 0.9668 0.9605 0.9178 0.9580 CSIQ AWGN 0.8974 0.9380 0.8431 0.9575 0.9541 0.9441 0.9359 0.9440 0.9670 0.9329 0.9652 JPEG 0.9543 0.9662 0.9412 0.9705 0.9615 0.9502 0.9664 0.9632 0.9689 0.9564 0.9573 JP2 K 0.9605 0.9683 0.9252 0.9672 0.9752 0.9643 0.9704 0.9648 0.9777 0.9630 0.9545 AGPN 0.8924 0.9059 0.8261 0.9511 0.9570 0.9357 0.9370 0.9387 0.9516 0.9517 0.9492 GB 0.9608 0.9782 0.9527 0.9745 0.9602 0.9643 0.9729 0.9589 0.9789 0.9664 0.9574 CTC 0.7925 0.9539 0.4873 0.9345 0.9207 0.9527 0.9438 0.9354 0.9324 0.9399 0.9273 LIVE JP2 K 0.9614 0.9649 0.9113 0.9696 0.9676 0.9323 0.9724 0.9700 0.9719 0.9608 0.9822 JPEG 0.9764 0.9808 0.9468 0.9846 0.9764 0.9584 0.9840 0.9778 0.9836 0.9674 0.9836 AWGN 0.9694 0.9667 0.9382 0.9858 0.9844 0.9799 0.9716 0.9774 0.9809 0.9457 0.9837 GB 0.9517 0.9720 0.9584 0.9728 0.9465 0.9066 0.9708 0.9518 0.9662 0.9561 0.9641 FF 0.9556 0.9442 0.9629 0.9650 0.9569 0.9237 0.9519 0.9402 0.9592 0.9627 0.9633 下载:
导出CSV
IQA模型 运行时间/s IQA模型 运行时间/s PSNR 0.0186 RFSIM 0.1043 SSIM 0.0892 FSIMc 0.3505 IW-SSIM 0.6424 GSM 0.1018 IFC 1.1554 CVSS 0.0558 VIF 1.1825 MPCC — MAD 2.7711 CAGS 0.4814 下载:
导出CSV
-
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34]
下载:
计量
- 文章访问数:7002
- PDF下载量:131
- 被引次数:0