Subjective And Objective Research On The Quality Of Non-reference Images For Screen Content

With the rapid development of multi-device interactive applications,the demand for transmission and processing of screen content image(SCI)is growing with each passing day.Different from traditional natural images,screen content images contain more media type and information.Screen content image usually contains image,graphic and text at the same time.The research of topic of image quality assessment is the basis of many applications,for example it can build the prior knowledge for development of video coding standard.So far,the research of topic of image quality assessment is mainly focused on traditional natural images.Owing to some properties that screen content images different from traditional natural images,the image quality assessment metrics that are designed for traditional natural images are hard to achieve same performance level when they are applied to screen content images.In order to promote screen content image related applications,the research of topic of image quality assessment specific to screen content image is becoming very important and urging.Standard image quality assessment database is the basis of proposal of objective image quality assessment metrics.Therefore,a large scale Immersive Media Laboratory screen content image quality database(IML-SCIQD)is first constructed by the use of Single stimulus continuous quality evaluation(SSCQE)subjective quality assessment method.The IMLSCIQD database contains 25 reference images and 1250 distorted images that are distorted by 10 distortions,each distortion with 5 distortion level.The selection of reference images ensure diversity of visual content and different media type layout style.The distortion type also meets the requirement of diversity.The IML-SCIQD database could be a good complement to the first large scale screen content image quality assessment database(SIQAD).At present,it has been presented that the visual perception properties that textual region and pictorial region of screen content images bring to human are different.In this paper,we further find that the statistical properties of these two different regions are also different even for the cases that screen content images are distorted.For single region,textual region or pictorial region,the statistical properties are also different when SCI is distorted by different distortion type.Based on the IML-SCIQD database and inspired by the idea of Natural Scene Statistics(NSS)based No Reference(NR)image quality assessment metrics,the Natural Scene Statistics based no reference screen content image quality assessment metric(NSNRS)is proposed.Screen content image is first segmented into textual region and pictorial region in the NSNRS metric.The quality scores of these two regions are then separately computed.At last,the quality score of whole screen content image is the combination of these two regions’ quality scores through weights.The NSNRS metric are at last compared with 12 state-of-the-art objective image quality assessment metrics on IML-SCIQD database and SIQAD database.Compared with the 5 stateof-the-art NSS-based No Reference metrics in these 12 metrics,The NSNRS metric has a great performance promotion.Compared with the other 7 state-of-the-art Full Reference and Reduced Reference metrics especially designed for traditional natural images and screen content images,the NSNRS metric has worse correlation result,our metric are No Reference metric after all.But for some distortion type,our metric even surpass some Full Reference metrics.