Petr Dostal

Image quality and QoE in multimedia systems

The image and video sensing and processing systems are of high importance nowadays. The crucial parameter is image quality as a part of QoS or more recently QoE (Quality of Experience) as a subjective impact (perception). This invited paper deals with various qualitative aspects in imaging systems. The first part is devoted to the modeling of HVS and image quality evaluation. A model for prediction of perceived quality of degraded images is presented. This model is optimized while incorporating visual attention coordinates. In the following parts four different examples of qualitative impacts (distortions and artifacts) of image post-processing are listed and shown. The first one is the image enhancement, second one image restoration by iterative detectors based on the maximum a posteriori probability criterion, third one image interpolation and finally fourth one image compression and transmission.

Selected problems of color image demosaicing in security imaging

The color TV security cameras are mostly based upon a single-sensor configuration with on-chip CFA (color filter array, Bayer configuration). The impact of different spatial sampling rasters in R G B channels causes an additional image distortion. In the field of security technology the identification task is among very frequent and above-mentioned sampling effect can affect it significantly. There are numerous interpolation algorithms suitable for the demosaicing - recalculation of full resolution R G B rasters. The paper is devoted to the comparison of several selected demosaicing algorithms usually applied in image enhancement of natural images where the subjective image quality is crucial. Our approach is oriented to the evaluation of identification limits in presence of noise and we are studying the noise resistivity of various demosaicing algorithms. We have chosen following demosaicing algorithms: Adaptive Homogeneity-Directed Demosaicing Algorithm and Demosaicing With Directional Filtering and a posteriori Decision and three different types of noise: Photon noise, Thermal noise and salt pepper. In order to evaluate the subjective image quality two approaches are used. The first approach is a standard subjective quality assessment we have applied the identification threshold evaluation. As the second one the objective quality metric of noisy image subjective quality SSIM, Structural SIMilarity index [1] is used. Both identification threshold and subjective quality dependencies are compared in order to evaluate a correlation degree. As test images we have chosen the car registration plate as an example of identification task esp. when the camera is recording images under low-light-level conditions. To keep the noise level precisely defined the test pictures were recorded almost noise-free and the noise was added consequently. The NEF (Nikon Electronic Format) picture format was applied to eliminate in-built image processing in camera.

Analysis of reconstruction performance of different demosaicing algorithms

Many proposed demosaicing algorithms used in single chip cameras are perceptually optimized only for multimedia imaging system where the QoE is used as criteria of quality. Concerning applied imaging system (e.g. security) the optimization criteria are totally different (e.g. character identification) and therefore the well rated algorithms may not be suitable for security. Four state-of-art interpolation techniques have been applied on the synthetic Snellen eye chart image test in order to evaluate the performance of character reconstruction. After that the image quality has been evaluated by the group of observers and two different objective quality metrics (PSNR, SSIM) have been calculated for comparison.

Advanced color image demosaicing in security imaging technology

Image records are of extreme importance in the security technology. Recent color photographic cameras are mostly constructed in a single-chip sensor configuration and the color splitting structure is designed as a planar filter array, therefore some interpolation technique has to be used to compute full-color image. The final image quality is affected by selected interpolation technique. The applied interpolating algorithms are perceptually optimized but the security technology criteria are different. Four state-of-art interpolation techniques have been applied on the selected set of security image tests. After that the image quality has been evaluated by the group of observers and two different objective quality metrics (PSNR, SSIM) have been calculated for comparison. As security image tests, a human face has been chosen and test target has been partially covered by distracters (sunblind) and recorded under complex illumination conditions. Furthermore the influence of SNR is discussed and demonstrated.

Modeling and evaluation of image quality in wireless surveillance networks

This paper studies the security image quality attained in streaming video over wireless networks under different packet error models. Wireless video sensor networks are becoming common place due to increasing crime rates and growing terrorist activities. A crucial requirement for video surveillance technology especially for real-time monitoring is that the video quality must not fall below a certain threshold so that objects and events in these videos could be identified and properly interpreted by viewers. However, the time varying transmission characteristics of the wireless channel and limited throughput can lead to poor performance of multimedia traffic over wireless networks. As a result, the end user may perceive jerky motions, frame freezes, and missing segments which may affect their ability to recognize objects and effectively analyze the scenes. An experimental set-up required for wireless network surveillance using IEEE 802. is described. The authors adopt subjective user tests to study the degree of identification of objects and events in security videos generated from camera security systems and subjected to wireless channel packet loss conditions to benchmark identification in the degraded sequences. Packet loss conditions were introduced into the sequences using NS-2 network simulator.

HLFSIM: Objective image quality metric based on ROI analysis

The image/video quality is a key issue in security video systems. Therefore the objective image/video quality criteria are extensively studied. In this paper, the novel full reference objective metric for image quality assessment is proposed. This metric is based on FSIM. The ROI detection is embedded in order to improve the performance. For ROI estimation, the ground truth data together with two different algorithms were used and compared - The security and multimedia images from LIVE database were used for performance evaluation. The correlation between the objective and subjective tests of multimedia images was calculated using Pearsons and Spearman Rank Order Correlation Coefficient. For performance comparison the state-of-the-art full reference objective image quality metrics were used; PSNR, SSIM, MS-SSIM, VIF and FSIM. In our previous paper, the importance of demosaicing technique and ROI has been shown. This paper continues in this topic and implements new full reference objective metrics for the reconstructed image quality evaluation. The results reveal that the ROI controlled by bottom-up mechanism can be used for performance improvement.

Cost-effective automatic stereoscopic security video system 3DSec

Object detecting and camera tracking in a large range of distances is still important problem in the research field of security and more recently also in the research field of assistive technologies too, when it is necessary to monitor intruders in the protected perimeter or elderly and disabled people both in hospital and at home. This paper discusses the design of an automatic stereoscopic video system using two identical pan-tilt-zoom (PTZ) cameras. A PTZ camera can provide a detailed live video of selected areas within a large potential field of view. Using two such cameras the TV system can be enriched by the possibility of estimating distances of monitored object or monitoring of two selected areas simultaneously. It means that both cameras can be jointly driven as a stereoscopic pair or can be run independently.

Locally adaptive demosaicing technique for security images based upon region-of-interest analysis

This paper presents the novel demosaicing approach that is based upon region of interest analysis. The question of if only the reconstruction quality of areas where human gazes during identification process of security objects are crucial for quality of identification is solved. Two fixation density maps were constructed on the basis of the ground through visual attention data which was acquired during an eye tracking experiment. Two security scenes containing traffic signs were used as testing images. SSIM and VIF with subjective testing were used for evaluation of reconstruction quality. The results tackle the questions — if quality of identification mainly depends on the quality of the areas where human eye is attracted, areas of interest, and whether the quality of the remaining part of image perceived by peripheral vision is crucial for quality of identification. Furthermore, the results show that embedding the information relating to these visual fixation patterns into demosaicing procedure can preserve the perceived image quality and quality of identification, while the computational complexity of proposed approach can be low.