An evaluation of the next-generation image coding standard AVIF Nabajeet Barman and Maria G. Martini School of Computer Science and Mathematics, Kingston University, London, UK {n.barman, m.martini}@kingston.ac.uk Abstract—This paper presents a comparative performance evaluation of the newly proposed AV1 Image File Format (AVIF) vs. other state-of-the art image codecs, for natural, synthetic and gaming images. The codecs are compared in terms of Rate-quality curves and BD-Rate savings considering different quality metrics. AVIF results in the best overall performance considering both 4:2:0 and 4:4:4 chroma sub- sampling encoded images for all types of images. Index Terms—AVIF, JPEG, Image Coding Standard, Syn- thetic Images, Gaming Images. I. I NTRODUCTION The human vision system (HVS) is highly responsive to visual aids, such as images and videos. Images today are used across a wide range of applications, from social media, such as Facebook, Instagram and Snapchat, to the field of machine learning and artificial intelligence for various tasks such as pattern recognition and detection of tumours in medical images. The recent advancements and acceptance of such applications are partly made possible due to the increased bandwidth availability and the improvement in image compression and processing capabilities. Image compression is a widely investigated and studied field, with many image compression standards being available for both lossy and lossless encoding with application-specific compression standards and techniques also being proposed [1] [2]. JPEG, which has been in use since 1992, is currently the most widely used lossy image compression standard especially for Internet applications and digital cameras, with almost 70% of websites using it now. Its successor, JPEG 2000 [3], is a discrete wavelet transform based compression standard shown to provide better image quality than JPEG and supporting both lossless and lossy image compression within the same file. WebP, developed in 2010 as a com- petitor of JPEG for use in web applications, is another image compression standard currently being developed by Google. Comparison studies presented in [4] have shown it to be 25-34% more efficient for the same SSIM value. The High Efficiency Image File Format (HEIF) is a standard that supports the storage of image data encoded using the HEVC standard and is shown to provide up to 25% reduction in bitrate compared to JPEG 2000 for the same objective quality [5]. Similar to HEIF, AVIF, which is the latest image compression standard, allows encapsulating AV1 intra-frame coded content and supports High Dynamic Range (HDR) and Wide Color Gamut (WCG) images as well as Standard Dynamic Range (SDR) [6]. So far, the newly developed AVIF format has been evaluated only on natural images by Netflix [7]. Hence, we present in this paper the first independent comparative evaluation of AVIF image coding format on three different datasets consisting of (a) Dataset 1 (b) Dataset 2 (c) Dataset 3 Fig. 1: Sample images from the three datasets. natural, gaming and synthetic images (see Section II-A). For further investigations and reproducibility of the results, we additionally provide the gaming images as an open-source dataset 1 . The rest of the paper is organized as follows. Sec- tion II presents the evaluation of dataset and methodology. Section III presents the results and observations of this study and Section IV concludes the paper with a discussion of future work. II. EVALUATION DATASET AND METHODOLOGY A. Evaluation Dataset To evaluate the performance of the newly proposed AVIF codec versus the existing codecs, we selected three different datasets. We report two sample images from each dataset in Figure 1. A description of the considered datasets is reported below. 1) Dataset 1 (D1): We used a total of 52 Images with resolution 2040 × 1346 from the DIV2K dataset, which consists of a wide range of natural images in .jpg format depicting real-world scenes such as monuments and landscape [8] and is similar to the ones used in [7]. 2) Dataset 2 (D2): Since the performance of quality as- sessment metrics is different for gaming content [9] as well as because gaming content is perceived differently from natural content [10], we use in this work a gaming 1 https://kingston.box.com/s/q6rsdzjg53ur61kqfve9vye1r3ovqtwg