Image Quality Assessment Using Similar Scene as Reference
Most of Image Quality Assessment (IQA) methods require the reference image to be pixel-wise aligned with the distorted image, and thus limiting the application of reference image based IQA methods. In this paper, we show that non-aligned image with simila
- PDF / 1,348,397 Bytes
- 16 Pages / 439.37 x 666.142 pts Page_size
- 49 Downloads / 293 Views
ract. Most of Image Quality Assessment (IQA) methods require the reference image to be pixel-wise aligned with the distorted image, and thus limiting the application of reference image based IQA methods. In this paper, we show that non-aligned image with similar scene could be well used for reference, using a proposed Dual-path deep Convolutional Neural Network (DCNN). Analysis indicates that the model captures the scene structural information and non-structural information “naturalness” between the pair for quality assessment. As shown in the experiments, our proposed DCNN model handles the IQA problem well. With an aligned reference image, our predictions outperform many state-of-the-art methods. And in more general case where the reference image contains the similar scene but is not aligned with the distorted one, DCNN could still achieve superior consistency with subjective evaluation than many existing methods that even use aligned reference images. Keywords: Image Quality Assessment · Similar scene referenced image · Structural similarity · “Naturalness” · Dual-path Deep Convolution Neural Network
1
Introduction
Assessing the quality of a distorted image would benefit from the availability of a reference image. As revealed in [1], human are more skilled at comparing images than making direct judgement of the image quality. Accordingly, human can evaluate the quality of an image more accurately and consistently when provided with a high-quality reference image, and meanwhile human may give different quality scores to the same image if different reference images are presented [2]. The situation is the same for Image Quality Assessment (IQA) algorithms, where methods that make use of reference images could achieve better consistency with Electronic supplementary material The online version of this chapter (doi:10. 1007/978-3-319-46454-1 1) contains supplementary material, which is available to authorized users. c Springer International Publishing AG 2016 B. Leibe et al. (Eds.): ECCV 2016, Part V, LNCS 9909, pp. 3–18, 2016. DOI: 10.1007/978-3-319-46454-1 1
4
Y. Liang et al.
Fig. 1. Using non-aligned images with similar scene as quality reference. Original images (top), distorted images (middle) and reference images (bottom) that are only similar to but is neither aligned with nor related by any geometrical transformation with the distorted images
subjective assessments than those that do not consider references [3,4]. Based on whether and how reference images are used, existing IQA methods could be broadly categorized into the following three groups: full reference (FR) IQA methods [2,3,5,6], reduced referenced (RR) IQA methods [7,8], and no reference (NR) IQA methods [9,10]. The former two groups, i.e.,the FR-IQA and the RR-IQA methods groups, take advantage of complete or partial information of the reference image respectively, while the NR-IQA methods are often designed to extract discriminative features [9,10] or to calculate natural scene statistics to qualify the image quality [11]. As explained above,
Data Loading...
