In addition to colors and shapes, factors of material appearance such as glossiness, translucency, and roughness are important for reproducing the realistic feeling of images. In general, these perceptual qualities are often degraded when reproduced as digital color images. Therefore, it is useful to enhance and reproduce them. In this article, the authors propose a material appearance enhancement algorithm for digital color images. First, they focus on the change of pupil behaviors, which is the first of the early vision systems to recognize visual information. According to their psychophysiological measurement of pupil size during material observation, they find that careful observation of surface appearance causes the pupil size to contract further. Next, they reflect this property in the retinal response, which is the next system in early vision. Then, they construct a material appearance enhancement algorithm named “PuRet” based on these physiological models of pupil and retina. By applying the PuRet algorithm to digital color test images, they confirm that perceived material appearance, including glossiness, transparency, and roughness, in the images is enhanced by using their PuRet algorithm. Furthermore, they show possibilities to apply their algorithm to a material appearance management system that could produce equivalent appearance qualities among different imaging devices by adjusting one parameter of PuRet.
Capsule endoscopy, using a wireless camera to capture the digestive track, is becoming a popular alternative to traditional colonoscopy. The images obtained from a capsule have lower quality compared to traditional colonoscopy, and high-quality images are required by medical doctors in order to set an accurate diagnosis. Over the last years several enhancement techniques have been proposed to improve the quality of capsule images. In order to verify that the capsule images have the required diagnostic quality some kind of quality assessment is required. In this work, the authors evaluate state-of-the-art no-reference image quality metrics for capsule video endoscopy. Furthermore, they use the best performing metric to optimize one of the capsule video endoscopy enhancement methods and validate through subjective experiment.
In this article, the authors propose a method for hiding a visual watermark in color printed images with arbitrary, natural content. The embedded watermark is imperceptible under normal illumination, but it is revealed under UV illumination. Their method is using the white-paper fluorescence as a source of the watermark signal. The binary visual watermark is embedded by controlling the amount of fluorescence during the halftoning process, which is achieved by modulating the fractional white area of exposed paper substrate. The method is based on the iterative Color Direct Binary Search halftoning, which ensures high quality of the printed images, and uses a suitable error metric to control the perceived distortion due to the watermark embedding. Results show that the proposed method achieves low perceptibility of the embedded watermark under normal illumination, while the watermark is easily detectable using a portable UV flashlight even in bright daylight conditions.
Visual perception of materials that make up objects has been gaining increasing interest. Most previous studies on visual material-category perception have used stimuli with rich information, e.g., color, shape, and texture. This article analyzes the image features of the material representations in Japanese “manga” comics, which are composed of line drawings and are typically printed in black and white. In this study, the authors first constructed a manga-material database by collecting 799 material images that gave consistent material impressions to observers. The manga-material data from the database were used to fully train “CaffeNet,” a convolutional neural network (CNN). Then, the authors visualized training-image patches corresponding to the top-n activations for filters in each convolution layer. From the filter visualization, they found that the filters reacted gradually to complicated features, moving from the input layer to the output layer. Some filters were constructed to represent specific features unique to manga comics. Furthermore, materials in natural photographic images were classified using the constructed CNN, and a modest classification accuracy of 63% was obtained. This result suggests that material-perception features for natural images remain in the manga line-drawing representations.
It is estimated that about 5–10% of the male population has some kind of color vision deficiency (CVD). For them, it is difficult or even impossible to distinguish certain colors. Many image enhancers exist, mostly based on hue changes, since CVDs are usually modeled at spectral level. In this article, the authors consider another point of view, investigating the role of luminance contrast to treat CVD.
In the following, the authors present a test, administered as a mobile application, to assess the performance of SiChaRDa, a recently proposed image enhancer, inspired by a model of the human visual system, that modifies the lightness of the image. The results indicate a role of contrast and edges in the readability of images for color vision-deficient people; however, they do not support a clear and unambiguous interpretation.
In this article, the authors propose a novel method of video magnification based on separation of the chromophore component. A video magnification method, Eulerian video magnification, was originally proposed by Wu et al. The method is effective in amplifying slight changes of facial color and can visualize the blood flow in the human face effectively. However, the conventional method was evaluated under stable conditions of illumination. It is necessary to enhance its robustness against environmental change for practical use. The proposed method amplifies the variation of the chromophore component which is separated from the shading component. The authors confirm that the proposed method can visualize the blood flow in the human face without artifacts caused by shading change. They also apply the video magnification framework to a tongue movie as preliminary work for medical application and confirm its effectiveness in visualizing the blood pulse and avoiding any clear artifacts.
Spectral filter array (SFA) technology requires development on demosaicing. The authors extend the linear minimum mean square error with neighborhood method to the spectral dimension. They demonstrate that the method is fast and general on Raw SFA images that span the visible and near infra-red part of the electromagnetic range. The method is quantitatively evaluated in simulation first, then the authors evaluate it on real data by the use of non-reference image quality metrics applied on each band. Resulting images show a much better reconstruction of text and high frequencies at the expense of a zipping effect, compared to the benchmark binary-tree method.
Dehazing methods based on prior assumptions derived from statistical image properties fail when these properties do not hold. This is most likely to happen when the scene contains large bright areas, such as snow and sky, due to the ambiguity between the airlight and the depth information. This is the case for the popular dehazing method Dark Channel Prior. In order to improve its performance, the authors propose to combine it with the recent multiscale STRESS, which serves to estimate Bright Channel Prior. Visual and quantitative evaluations show that this method outperforms Dark Channel Prior and competes with the most robust dehazing methods, since it separates bright and dark areas and therefore reduces the color cast in very bright regions.
Image enhancement using visible (RGB) and near-infrared (NIR) image data has been shown to enhance useful details of the image. While the enhanced images are commonly evaluated by observers’ perception, in the present work, we rather evaluate it by quantitative feature evaluation. The proposed algorithm presents a new method to enhance the visible images using NIR information via edge-preserving filters, and also investigates which method performs best from an image features standpoint. In this work, we combine two edge-preserving filters: bilateral filter (BF) and weighted least squares optimization framework (WLS). To fuse the RGB and NIR images, we obtain the base and detail images for both filters. The NIR-detail images for both filters are simply fused by taking an average/maximum of both, which is then combined with the RGB-base image from the WLS filter to reconstruct the final enhanced RGB-NIR image. We then show that our proposed enhancement method produces more stable features than the existing state-of-the-art methods on RGB-NIR Scene Dataset. For feature matching, we use the SIFT features. As a use case, the proposed fusion method is tested on two challenging biometric verifications tasks using CMU hyperspectral face and CASIA multispectral palmprint databases. Our exhaustive experiments show that the proposed fusion method performs equally well in comparison to the existing biometric fusion methods.