Turkish Journal of Electrical Engineering and Computer Sciences
Author ORCID Identifier
MENGJUAN ZHAO: 0009-0007-1180-2673
YITAO LIANG: 0009-0009-3501-502X
WEIYA SHI: 0000-0001-9116-7305
JUAN XIA: 0000-0001-8391-5262
Abstract
Currently, grayscale images are preferred as input data for some specific vision tasks. Decolorization is the transformation of a color image into a grayscale image. Efficient decolorization algorithms can improve the overall task efficiency, while perceptual preservation in decolorization can provide more information for further processing. In recent research, traditional methods focus on preserving contrast or detail information with little attention to perceptual features. Deep-learning methods are beginning to consider perceptual preservation, but they run inefficiently. In addition, the decolorization methods lack the optimal target grayscale images for reference. Therefore, we propose a new deep learning-based real-time no-reference decolorization network with perceptual preservation, which includes a grayscale network and a perception enhancement network. The former learns the contrast and detail information of the image, and the latter is used to fit perceptual enhancement curves to enhance perceptual preservation. Meanwhile, a new set of no-reference loss functions is formulated to evaluate the perceptual enhancement level and the decolorization quality. Experiments show that our approach can preserve the contrast, detail, and perceptual information in the image well and achieve good performance in the existing grayscale quality evaluation metrics. The method maintains real-time performance, requiring only 3.5 ms to process a 256×256 color image, making it ideal for devices with real time requirements and output limitations. Our approach achieves a good balance between algorithmic efficiency and effectiveness.
DOI
10.55730/1300-0632.4161
Keywords
Image decolorization, no-reference learning, convolutional neural network, real-time, perceptual preservation
First Page
31
Last Page
48
Publisher
The Scientific and Technological Research Council of Türkiye (TÜBİTAK)
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Recommended Citation
ZHAO, M, LIANG, Y, SHI, W, & XIA, J (2026). A deep learning-based real-time no-reference image decolorization network with perceptual preservation. Turkish Journal of Electrical Engineering and Computer Sciences 34 (1): 31-48. https://doi.org/10.55730/1300-0632.4161
Included in
Computer Engineering Commons, Computer Sciences Commons, Electrical and Computer Engineering Commons
