No abstract

An update on the JPEG XL standardization effort: JPEG XL is a practical approach focused on scalable web distribution and efficient compression of high-quality images. It will provide various benefits compared to existing image formats: significantly smaller size at equivalent subjective quality; fast, parallelizable decoding and encoding configurations; features such as progressive, lossless, animation, and reversible transcoding of existing JPEG; support for high-quality applications including wide gamut, higher resolution/bit depth/dynamic range, and visually lossless coding. Additionally, a royalty-free baseline is an important goal. The JPEG XL architecture is traditional block-transform coding with upgrades to each component. We describe these components and analyze decoded image quality.

No abstract

JPEG XL is a practical, royalty-free codec for scalable web distribution and efficient compression of high-quality photographs. It also includes previews, progressiveness, animation, transparency, high dynamic range, wide color gamut, and high bit depth. Users expect faithful reproductions of ever higher-resolution images. Experiments performed during standardization have shown the feasibility of economical storage without perceptible quality loss, lossless transcoding of existing JPEG, and fast software encoders and decoders. We analyse the results of subjective and objective evaluations versus existing codecs including HEVC and JPEG. New image codecs have to co-exist with their previous generations for several years. JPEG XL is unique in providing value for both existing JPEGs as well as new users. It includes coding tools to reduce the transmission and storage costs of JPEG by 22% while allowing byte-for-byte exact reconstruction of the original JPEG. Avoiding transcoding and additional artifacts helps to preserve our digital heritage. Applications require fast and low-power decoding. JPEG XL was designed to benefit from multicore and SIMD, and actually decodes faster than JPEG. We report the resulting speeds on ARM and x86 CPUs. To enable reproduction of these results, we open sourced the JPEG XL software in 2019.

The interpretation of satellite imagery benefits from merging the spatial structure of the high-resolution panchromatic image with the spectral information. Such "pan-sharpening" has been the topic of extensive research. One objective of our investigations is to process satellite images within seconds. In this work, we build upon the "Fast IHS" technique, using a weighted linear combination of the up-sampled multispectral bands to derive a composite image closer to what the panchromatic sensor had seen. The difference to the actual panchromatic image approximates the high-frequency detail signal and is added to the multispectral bands. However, fixed band weights (exemplified by the "Modified IHS" algorithm) cannot account for differing radiometry and atmospheric conditions. To further reduce color distortion, we compute the optimal band weights for a given data set in the sense of minimizing the mean-square difference between the composite and panchromatic images. Since the noise in the panchromatic image (sometimes non-linear) impacts a subsequent graphbased segmentation algorithm, an additional denoising step is applied before fusion. We use an improved approximation of the Bilateral Filter, which preserves edges and requires only one fast iteration. The quality of the fused image is evaluated in a comparative study of pan-sharpening algorithms available in ERDAS IMAGINE 9.3. Objective metrics such as Q4 show an improvement in terms of color fidelity. The image segmentation results also demonstrate the applicability of this method towards automated image analysis.

Recent works showed that implementations of quicksort using vector CPU instructions can outperform the non‐vectorized algorithms in widespread use. However, these implementations are typically single‐threaded, implemented for a particular instruction set, and restricted to a small set of key types. We lift these three restrictions: our proposed vqsort algorithm integrates into the state‐of‐the‐art parallel sorter ips4o$$ ip{s}^4o $$, with a geometric mean speedup of 1.59. The same implementation works on seven instruction sets (including SVE and RISC‐V V) across four platforms. It also supports floating‐point and 16–128 bit integer keys. To the best of our knowledge, this is the fastest sort for large arrays of non‐tuple keys on CPUs, up to 20 times as fast as the sorting algorithms implemented in standard libraries. This article focuses on the practical engineering aspects enabling the speed and portability, which we have not yet seen demonstrated for a quicksort implementation. Furthermore, we introduce compact and transpose‐free sorting networks for in‐register sorting of small arrays, and a vector‐friendly pivot sampling strategy that is robust against adversarial input.

We report on pairwise comparisons by human raters of JPEG images from libjpeg and our new Guetzli encoder. Although both files are size-matched, 75% of ratings are in favor of Guetzli. This implies the Butteraugli psychovisual image similarity metric which guides Guetzli is reasonably close to human perception at high quality levels. We provide access to the raw ratings and source images for further analysis and study.

This report introduces a new lossless asymmetric single instruction multiple data codec designed for extremely efficient decompression of large satellite images. A throughput in excess of 3GB/s allows decompression to proceed in parallel with asynchronous transfers from fast block devices such as disk arrays. This is made possible by a simple and fast single instruction multiple data entropy coder that removes leading null bits. Our main contribution is a new approach for vectorized prediction and encoding. Unlike previous approaches that treat the entropy coder as a black box, we account for its properties in the design of the predictor. The resulting compressed stream is 1.2 to 1.5 times as large as JPEG-2000, but can be decompressed 100 times as quickly - even faster than copying uncompressed data in memory. Applications include streaming decompression for out of core visualization. To the best of our knowledge, this is the first entirely vectorized algorithm for lossless compression