Abstract—Traditional filters based on full window consider that neighbor pixels surround target pixel, which likely crosses over the potential edges. Although the side win dow (SW) with eight sub-windows alleviates artifacts and halos, it still faces a challenge with mandatory smoothing. To address the issue, a novel and simple steerable side window (SSW) framework is proposed in the paper. The proposed SSW utilizes derivation of Gaussian function to generate a set of side windows including half- and quarter-window with arbitrary orientation, which effectively improves the perception of detail information and significantly enhances adher ence to the pixel distribution. Meanwhile, the SSW can also be easily embedded in a variety of filters. It not only inherits the properties of SW, but also extends the flexibility of SW. Experiments are carried out on different filters, and the results show that the proposed SSW technique is both effective and efficient, enabling more advantageous applications.
Fig.1 The architecture of SSW.A set of dedicated filters with different orientations handles the input image.The filtered image is adaptively constructed from input image and a set of latent maps.
Fig. 1 shows the architecture of SSW. The proposed SSW utilizes the basis filters to convolute the test image. Then it selects the smallest deviation between the input image and latent maps to construct the final output. The SSW incorporates more side local information from the image, so it is easier to obtain high-quality filtering results.
✨ Please cite us if this work is helpful to you
@ARTICLE{
author={Xiaohong Jia, Tao Lei, Yingbo Wang, Xuejun Zhang, Guanghui Yan and Asoke K. Nandi},
journal={Neural Computing and Applications},
year={2026}
volume={38}
article number={80}
title={The Design and Application of Steerable Side Window Framework},
keywords={Steerable side window, Arbitrary orientation, Smoothing, Edge preserving},
doi={https://doi.org/10.1007/s00521-025-11819-w}