Subjective assessment of color night vision fusion image quality for typical scenes
[Objective]Evaluating the quality of infrared and visible(microlight)color fusion images in specific scenes is important for developing color night vision fusion technology and embedded systems.Among these,subjective quality evaluation has emerged as a key area of research because it can yield results that align with human visual perception.However,current subjective evaluation methods are affected by the evaluation methodology and data quality,necessitating further investigation for specific applications.To address this,we have proposed a subjective evaluation method that combines dynamic referencing and multistimulus comparison.This approach is grounded in large-scale experiments conducted on the subjective evaluation of color-fused images from typical scenes.The outcome of these experiments is a subjective quality scoring database containing 1 648 color-fused images.In addition,we suggest a fuzzy comprehensive evaluation method,developed from experimental characterizations,to analyze the scene applicability of different fusion algorithms.[Methods]We propose a method that integrates dynamic reference stimuli and multistimulus comparison,tailored for evaluating typical application scenarios of fused images such as seascapes,skylines,urban areas,and forests.First,an expert creates a reference sequence based on subjective experience and assigns scores to the images using a dual-stimulus approach.This process aids observers in effectively comparing and analyzing the correlation between images.During the formal evaluation phase,the reference sequence and the image undergoing evaluation are displayed on two separate screens.We propose a method that leverages the concept of bubble sorting to initially sort the frame images on the same page,thereby reducing the number of comparisons in the evaluation process.We then assign 0-100 scores to the images,with the aim of assigning a more accurate quality score.At the same time,considering the key indicators for evaluating the quality of fused images,we propose to measure the visual quality from three different aspects:"target background contrast""color coordination"and"image clarity."Finally,we statistically process the subjective evaluation results and establish a quality subjective evaluation database containing 1 648 color fusion images.Considering the prominent fuzzy characteristics in the subjective quality evaluation results of fused images,we utilize the fuzzy comprehensive evaluation method to analyze the scene applicability of different fusion algorithms.[Results]The experimental results of the subjective evaluation method combining dynamic reference and multistimulus comparison reveal the following:1)The proposed evaluation method somewhat reduces the large number of comparison times and time required by the forced choice pairwise comparison method and reduces the uncertainty in the evaluation of the single-stimulus method.2)In the consistency analysis of the experimental data,a Pearson's correlation coefficient is observed,suggesting that the data set developed in this paper exhibits strong validity and reliability.3)Utilizing the comprehensive quality rating scale of typical scene fusion images generated by the fuzzy comprehensive evaluation method based on experimental characteristics,we compare and analyze the original image.The outcome reveals an assessed quality rating that aligns more closely with the subjective perception of the human eye.[Conclusions]The proposed subjective evaluation method combining dynamic reference and multistimulus comparison effectively improves the accuracy of subjective evaluation scores.Meanwhile,the establishment of an image subjective evaluation database of typical scenes provides a solid experimental and theoretical foundation for conducting research on the evaluation of night vision fusion image quality.
color fusion imagessubjective quality assessmentmultistimulus comparison methodtypical scenariosfuzzy comprehensive evaluation
NETL
NSTL
万方数据
