http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
초협대역 전송 시스템상에서 H.265/HEVC 부호화 저해상도 비디오에 대한 주관적 화질 평가
에이. 에프. 엠. 샤합 우딘,엠에스티. 시라잠 모니라,정태충,김동현,최증원,전기남,배성호 한국방송∙미디어공학회 2019 방송공학회논문지 Vol.24 No.6
In this paper, we perform a subjective quality assessment on low-resolution surveillance videos, which are encoded with a very low target bit-rate to use in an ultra-low band transmission system and investigate the encoding effects on the perceived video quality. The test videos are collected based on their spatial and temporal characteristics which affect the perceived quality. H.265/HEVC encoder is used to prepare the impaired sequences for three target bit-rates 20, 45, and 65 kbps and subjective quality assessment is conducted to evaluate the quality from a viewing distance of 3H. The experimental results show that the quality of encoded videos, even at target bit-rate of 45 kbps can satisfy the users. Also we compare objective image/video quality assessment methods on the proposed dataset to measure their correlation with subjective scores. The experimental results show that the existing methods poorly performed, that indicates the need for a better quality assessment method.
에이. 에프. 엠. 샤합 우딘,김동현,엠에스티. 시라잠 모니라,최증원,정태충,배성호 한국방송∙미디어공학회 2023 방송공학회논문지 Vol.28 No.6
Video quality assessment has become an essential step in many image processing and computer vision applications, especially whenthe task is to maintain the quality of service where the video is transmitted from source to destination. In real world, there are somespecialized communication networks e.g., military communication, where the bandwidth is extremely limited. Also, the characteristicsof the videos that are transmitted in such ultra-narrow band transmission systems, are usually different from general-purpose videos. Specifically, these videos are highly concentrated on some specific objects or regions. Consequently, the existing image or videoquality assessment methods are not suitable to predict the quality of this kind of saliency videos and also there is no related datasetavailable. In this paper, we first propose a video quality assessment dataset that contains carefully collected extremely compressedsurveillance videos along with their visual qualities that are evaluated by human. Then we also propose a deep neural network as afull-reference video quality assessment method. Since there are no patch-wise ground truth quality values, the proposed method passesthe whole video and its corresponding saliency map to the convolutional neural network to extract the features which is thentransferred to the regressor to estimate the final quality score. Instead of patch-wise feature extraction followed by an error poolinglayer to predict the overall quality, the proposed method makes use of the full image signal to extract important features and predictthe overall quality. This approach helps to achieve a higher receptive field and does not require any error pooling stage. Theextensive experimental results validate the prediction performance of the proposed quality assessment method.