This study aims at developing an image quality metric for exposure quality, with a transform to just noticeable differences (JNDs) of quality in pictorial scenes. Such a perceptually calibrated exposure metric would allow the prediction of overall image quality by combining exposure with other image attributes. Eight pictorial images were used in the study, and twenty-one observers participated in the subjective evaluation using a softcopy quality ruler method defined in ISO 20462 Part 3. The image simulation path involved seven levels of exposure manipulation, together with two variations in tone mapping algorithms (a global tone mapping algorithm and a local tone mapping algorithm). For each pictorial scene a second image was captured with an exposure target in the scene, allowing the measurement of the scene exposure level. The results showed that an objective metric based on the green channel intensity of the exposure target could be used to predict the optimal exposure level and the quality falloff due to exposure error.
Having a methodology for assessing smartphone camera image quality is advantageous for both those who design and develop the cameras as well as those who use them. Camera engineers need to quickly and reliably assess the impact of the system decisions they make. Smartphone customers who are armed with a quantitative understanding of the image quality can include this information to make informed decisions between products. This research project was undertaken to develop a procedure for evaluating pictorial image quality for smartphone camera captures. Experiments were conducted to evaluate tone quality, color quality, and sharpness and noise using images captured with 20 cameras that were released primarily in the period between 2012 to late 2014. A variety of scenes were captured with each device. In each test, observers rated the test images for overall quality and then for a specific image quality characteristic using an anchored scaling experimental protocol. The results indicated high correlations between the individual characteristics and overall quality. It was also determined that high correlations could be achieved between the visual results and objective measurements for sharpness and noise. Both analyses indicated that a two-step process in which devices are first sorted into categories of high and low quality followed by a second sort to further refine device quality may be required to successfully predict the visual results.
Pixel blooming in CMOS image sensors is often characterized by metrics that are extracted from the photo-response curve. This work explores whether these pixel blooming metrics are in any significant correlation with end-user image quality in mobile photography, and provides quantitative answers to that question by studying changes in size and color of saturated image regions.
Multi-camera systems are increasingly gaining popularity for various applications and their correct functionality depends on precise registration. The complexity of registering the various images to each other is reduced significantly by rectifying the images. This usually relies on an offline calibration process. In reality, components of the camera module respond differently to various factors such as temperature variations, field conditions, etc. Therefore, changes in geometric camera calibration, unless accounted for, can affect the proper registration, which in turn leads to severe degradation of the imaging system or can lead to artifacts. We present a method that can assess the geometric calibration of an array camera and perform an adaptive adjustment of geometric calibration by robust feature matching in any imaged scene. Assuming a gradual degradation of geometric calibration from their previously calibrated values, we exploit the redundancy of a camera array system to recover from the variation of calibrated parameters. Compared to other online calibration methods mostly used for stereo systems, our proposed method is efficient and robust, and derives a solution for multi-camera systems. We illustrate the usefulness of our geometric calibration compensation approach through a super-resolution application where we recover significant image details that are lost due to errors in calibration.
In this paper, we propose a system to automatically design image filters, for manufacturers of image capture devices to maintain desired image quality. The proposed system is based on measuring the Spatial Frequency Response (SFR) of the device using the slanted edge technique. This includes an automatic approach to crop the slanted edges and perform the measurements. Based on the measured SFR, an equalizing filter is automatically designed for the device to standardize its SFR to meet a certain goal, for example, to provide unity gain for low and middle frequency ranges while attenuating higher frequencies. In this way, different devices can share an equivalent frequency response and thus offer consistent image quality. A set of device-independent filters may then be cascaded with the equalizing filter of each device. These device-independent filters are designed once, while the numerous individual device-dependent filters are designed automatically. This procedure saves significant effort designing a large collection of individual filters, while improving the consistency of image quality across different image capture devices. To accommodate SFR variation after manufacturing, an end user could apply this approach, if embedded within the device.
The standard ISO12233:2014 describes different methods on how to measure the spatial frequency response (SFR) of an imaging system. It uses either a slanted edge (eSFR) or a harmonic Siemens star (sSFR). Within the document, it is mentioned that a linearization process shall correct a non-linear tone curve. Normalization is not further defined, but is an important part of the evaluation for the sSFR method. Using the sSFR method for texture loss analysis in the upcoming ISO19567 standard (based on a low contrast version of the Siemens star) it is even more critical. In this paper, we evaluate the influence of linearization and normalization on the results, identify issues in common implementations and present a new approach.
There has been a growing interest in recent years in the development of objective image quality assessment (IQA) models, whose roles are not only to monitor image quality degradations and benchmark image processing systems, but also to optimize various image and video processing algorithms and systems. While the past achievement is worth celebrating, a number of major challenges remain when we apply existing IQA models in realworld applications. These include obvious ones such as the challenges to largely reduce the complexity of existing IQA algorithms and to make them easy-to-use and easy-to-understand. There are also challenges regarding the applicability of existing IQA models in many real-world problems where image quality needs to be evaluated and compared across dimensionality, across viewing environment, and across the form of representations – specific examples include quality assessment for image resizing, color-togray image conversion, multi-exposure image fusion, image retargeting, and high dynamic range image tone mapping. Here we will first elaborate these challenges, and then concentrate on a specific one, namely the generalization challenge, which we believe is a more fundamental issue in the development, validation and application of IQA models. Specifically, the challenge is about the generalization capability of existing IQA models, which achieve superior quality prediction performance in lab testing environment using a limited number of subject-rated test images, but the performance may not extend to the real-world where we are working with images of a much greater diversity in terms of content and complexity. We will discuss some principle ideas and related work that might help us meet the challenges in the future.
Objective video quality metrics are designed to estimate the quality of experience of the end user. However, these objective metrics are usually validated with video streams degraded under common distortion types. In the presented work, we analyze the performance of published and known full-reference and noreference quality metrics in estimating the perceived quality of adaptive bit-rate video streams knowingly out of scope. Experimental results indicate not surprisingly that state of the art objective quality metrics overlook the perceived degradations in the adaptive video streams and perform poorly in estimating the subjective quality results.