Image quality assessment (IQA) is a fundamental metric for image processing tasks (e.g., compression). With full-reference IQAs, traditional IQAs, such as PSNR and SSIM, have been used. Recently, IQAs based on deep neural networks (deep IQAs), such as LPIPS and DISTS, have also been used. It is known that image scaling is inconsistent among deep IQAs, as some perform down-scaling as pre-processing, whereas others instead use the original image size. In this paper, we show that the image scale is an influential factor that affects deep IQA performance. We comprehensively evaluate four deep IQAs on the same five datasets, and the experimental results show that image scale significantly influences IQA performance. We found that the most appropriate image scale is often neither the default nor the original size, and the choice differs depending on the methods and datasets used. We visualized the stability and found that PieAPP is the most stable among the four deep IQAs.

Image quality assessment (IQA) is an inherent problem in the field of image processing. Recently, deep learning-based image quality assessment has attracted increased attention, owing to its high prediction accuracy. In this paper, we propose a fully-blind and fast image quality predictor (FFIQP) using convolutional neural networks including two strategies. First, we propose a distortion clustering strategy based on the distribution function of intermediate-layer results in the convolutional neural network (CNN) to make IQA fully blind. Second, by analyzing the relationship between image saliency information and CNN prediction error, we utilize a pre-saliency map to skip the non-salient patches for IQA acceleration. Experimental results verify that our method can achieve the high accuracy (0.978) with subjective quality scores, outperforming existing IQA methods. Moreover, the proposed method is highly computationally appealing, achieving flexible complexity performance by assigning different thresholds in the saliency map.

Inverse synthetic aperture radar (ISAR) is useful for automatic target recognition (ATR) because it can reconstruct a high resolution image of an observed target. In ISAR imaging, 3-dimensional reflectivity distribution of a target is projected to the plane defined by range axis and cross range axis. In order to recognize the observed target by using pattern matching, reference images of candidate targets must be adequately generated. However, that is difficult because the cross range axis, which depends on the target's unknown rotational motion, can not be determined precisely. In this paper, we propose a new algorithm to generate reference ISAR images of ship targets. In this algorithm, tracking data, Doppler width and the slope of the centerline of an ISAR target image are used to specify the cross range axis. The effectiveness of the proposed algorithm was evaluated by using simulated targets.

We propose a sea surveillance system that automatically detects intruding objects in the sea. The difficulty with an automatic system is detecting objects such as moving boats while reducing false positives caused by some waves and reflections in the sea. A false positive is reporting an object which doesn't actually exist, while a false negative is a failure in detecting an intruding object. Firstly, we identify factors of false positives. Secondly, we propose a new surveillance system considering these factors. Our proposed system combines three detecting methods. The first method is detection of Differences between Surveillance images and Flapping Reference images (DSFR). The second method is detection of Contours from Averaging images (CA). The third method is Silhouette object Detection (SD). The combination of DSFR and CA detects various moving objects under normal light conditions, while SD detects objects under backlight conditions. Finally we apply our proposed method to actual situations. Our proposed method detected boats while reducing false positives effectively.