Abstract
Conventional image compression techniques targeted for the perceptual quality are not generally optimized for classification tasks using deep neural networks (DNNs). To compress images for DNN inference tasks, recent studies have proposed task-centric image compression methods with quantization techniques optimized for DNN inference. Among them, color quantization was proposed to reduce the amount of data per pixel by limiting the number of distinct colors (color space) in an image. However, quantizing images into various color space sizes requires training and inference of multiple DNNs, each of which is dedicated to each color space. To overcome this limitation, we propose a scalable color quantization method, where images with variable color space sizes can be extracted from a master image generated by a single DNN model. This scalability is enabled by weighted color grouping that constructs a color palette using critical color components for the classification task. We also propose an adaptive training method that can jointly optimize images with various color-space sizes. The results show that the proposed method supports dynamic changes of the color space size between 1–6 bit color space per pixel, while even increasing the inference accuracy at a low bit precision up to 20.2% and 46.6% compared to other task- and human-centric color quantizations, respectively.
- [1] . 2016. Color quantization. In Digital Image Processing. Springer, 329–339.Google Scholar
Cross Ref
- [2] . 2021. End-to-end optimized image compression for machines, a study. In Data Compression Conference (DCC). IEEE, 163–172.Google Scholar
Cross Ref
- [3] . 2020. Task-aware quantization network for JPEG image compression. In European Conference on Computer Vision. Springer, 309–324.Google Scholar
Digital Library
- [4] . 2011. An analysis of single-layer networks in unsupervised feature learning. In 14th International Conference on Artificial Intelligence and Statistics. 215–223.Google Scholar
- [5] . 1999. Peer group filtering and perceptual color image quantization. In IEEE International Symposium on Circuits and Systems (ISCAS). IEEE, 21–24.Google Scholar
- [6] . 2001. Unsupervised segmentation of color-texture regions in images and video. IEEE Trans. Pattern Anal. Mach. Intell. 23, 8 (2001), 800–810.Google Scholar
Digital Library
- [7] . 2021. Towards task-generic image compression: A study of semantics-oriented metrics. IEEE Trans. Multim. (2021).Google Scholar
Digital Library
- [8] . 1988. A simple method for color quantization: Octree quantization. In New Trends in Computer Graphics. Springer, 219–231.Google Scholar
Cross Ref
- [9] . 1991. Colour image quantization for high resolution graphics display. Image. Vis. Comput. 9, 5 (1991), 303–312.Google Scholar
Cross Ref
- [10] . 1985. Clustering to minimize the maximum intercluster distance. Theoret. Comput. Sci. 38 (1985), 293–306.Google Scholar
Cross Ref
- [11] . 2016. Deep residual learning for image recognition. In IEEE Conference on Computer Vision and Pattern Recognition. 770–778.Google Scholar
Cross Ref
- [12] . 1982. Color image quantization for frame buffer display. ACM Siggraph Comput. Graph. 16, 3 (1982), 297–307.Google Scholar
Digital Library
- [13] . 2015. Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531 (2015).Google Scholar
- [14] . 2020. Learning to structure an image with few colors. In IEEE/CVF Conference on Computer Vision and Pattern Recognition. 10116–10125.Google Scholar
Cross Ref
- [15] . 1986. Quantization of color images for display on graphics terminals. In IEEE Global Telecommunications Conference. 1138–1142.Google Scholar
- [16] . 2015. Batch normalization: Accelerating deep network training by reducing internal covariate shift. arXiv preprint arXiv:1502.03167 (2015).Google Scholar
- [17] . 1952. Stochastic estimation of the maximum of a regression function. Ann. Math. Statist. 23, 3 (1952), 462–466.Google Scholar
- [18] . 2021. Target-dependent scalable image compression using a reconfigurable recurrent neural network. IEEE Access 9 (2021), 119418–119429.Google Scholar
- [19] . 2009. Learning multiple layers of features from tiny images. https://www.cs.toronto.edu/kriz/learning-features-2009-TR.pdf.Google Scholar
- [20] . 2015. Tiny ImageNet visual recognition challenge. CS 231N 7 (2015).Google Scholar
- [21] . 2018. DeepN-JPEG: A deep neural network favorable JPEG-based image compression framework. In 55th Annual Design Automation Conference. 1–6.Google Scholar
Digital Library
- [22] . 2016. SGDR: Stochastic gradient descent with warm restarts. arXiv preprint arXiv:1608.03983 (2016).Google Scholar
- [23] . 2022. Coding of Still Pictures - ds.jpeg.org. Retrieved from https://ds.jpeg.org/documents/jpegai/wg1n90021-REQ-JPEG_AI_Use_Cases_and_Requirements.pdf.Google Scholar
- [24] . 2015. Deep neural networks are easily fooled: High confidence predictions for unrecognizable images. In IEEE Conference on Computer Vision and Pattern Recognition. 427–436.Google Scholar
Cross Ref
- [25] . 1991. Color quantization of images. IEEE Trans. Sig. Process. 39, 12 (1991), 2677–2690.Google Scholar
Digital Library
- [26] . 2020. Semantic-preserving image compression. In IEEE International Conference on Image Processing (ICIP). IEEE, 1281–1285.Google Scholar
Cross Ref
- [27] . 2015. U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical Image Computing and Computer-assisted Intervention. Springer, 234–241.Google Scholar
Cross Ref
- [28] . 2017. Grad-Cam: Visual explanations from deep networks via gradient-based localization. In IEEE International Conference on Computer Vision. 618–626.Google Scholar
Cross Ref
- [29] . 2014. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556 (2014).Google Scholar
- [30] . 2019. Super-convergence: Very fast training of neural networks using large learning rates. In Artificial Intelligence and Machine Learning for Multi-Domain Operations Applications, Vol. 11006. International Society for Optics and Photonics, 1100612.Google Scholar
- [31] . 2020. Recognition-driven compressed image generation using semantic-prior information. IEEE Sig. Process. Lett. 27 (2020), 1150–1154.Google Scholar
Cross Ref
- [32] . 2019. Lossy image compression with recurrent neural networks: From human perceived visual quality to classification accuracy. arXiv preprint arXiv:1910.03472 (2019).Google Scholar
- [33] . 2021. Observer dependent lossy image compression. In Pattern Recognition: 42nd DAGM German Conference, DAGM GCPR’20, Tübingen, Germany, September 28–October 1, 2020, Proceedings 42. Springer, 130–144.Google Scholar
Digital Library
- [34] . 1992. Color quantization by dynamic programming and principal analysis. ACM Trans. Graph. 11, 4 (1992), 348–372.Google Scholar
Digital Library
- [35] . 1997. Color image quantization by minimizing the maximum intercluster distance. ACM Trans. Graph. 16, 3 (1997), 260–276.Google Scholar
Digital Library
- [36] . 2019. Any-precision deep neural networks. arXiv preprint arXiv:1911.07346 (2019).Google Scholar
- [37] . 2019. Universally slimmable networks and improved training techniques. In IEEE/CVF International Conference on Computer Vision. 1803–1811.Google Scholar
Cross Ref
- [38] . 2020. Precision gating: Improving neural network efficiency with dynamic dual-precision activations. arXiv preprint arXiv:2002.07136 (2020).Google Scholar
Index Terms
Scalable Color Quantization for Task-centric Image Compression
Recommendations
DWT-based scene-adaptive color quantization
Special issue on multi-dimensional image processingColor quantization is the process of reducing the number of colors in an image. That is, color quantization maps a large number of colors into a much smaller number of representative colors while keeping color distortion to an acceptable level. The ...
Comparative analysis of the quantization of color spaces on the basis of the CIELAB color-difference formula
This article discusses the CIELAB color spave within the limits of optimal colors including the complete volume of object colors. A graphical representation of this color space is composed of planes of constant lightness L* with an net of lines parallel ...
Color image quantization with peak-picking and color space
AbstractColor image quantization is a significant procedure of reducing the huge range of color values of a digital color image into a limited range. In this paper, an automated clustering of pixels and color quantization algorithm is proposed. The ideal ...






Comments