DCT and high-boost filtering improve grayscale multi-focus fusion in four-dataset study

A hybrid multi-focus image-fusion framework that combines high-boost filtering with discrete cosine transformation produced the strongest reported results among the tested approaches in experiments on four grayscale image datasets. The proposed method, called Multi-Focus Sharpened Image Fusion (MFSIF), aims to preserve fine textures, clearer boundaries and high-frequency detail when images of a scene with different areas in focus are combined. Its highest-performing configuration, labeled “DCT + HB,” achieved a peak signal-to-noise ratio of 39.5911, a root mean square error of 9.4122, a mean absolute error of 1.6306 and an entropy value of 0.9324. The study’s abstract reports that these measurements indicated better image quality and detail representation, although it does not provide uncertainty estimates for the reported values or the number of individual images tested.

Multi-focus image fusion creates one specified image from multiple images containing useful information in different areas of focus. The source material describes the approach as combining photographs of the same scene captured at different focal lengths, with the aim of maintaining sharpness across the scene’s depth range. Potential application areas include medical imaging, photography, surveillance and remote sensing. MFSIF was designed to address limitations attributed to some existing fusion approaches, including blurred boundaries, artifact generation and the loss of fine structural details, particularly in highly textured or defocused regions. Its empirical evaluation used four grayscale datasets named “Bottle,” “Clock,” “Leaves” and “Balloons,” so the directly reported findings are confined to those test collections rather than to color images or application-specific clinical and surveillance data.

The framework has a two-stage design. In the first stage, unsharp masking and high-boost filtering are used to improve the sharpness, contrast and quality of the input images. These edge-aware sharpening filters are intended to strengthen the representation of fine details before the images are merged. In the second stage, the enhanced images are combined using transform-domain techniques that include the Discrete Wavelet Transform (DWT) and the Discrete Cosine Transform (DCT). By integrating established sharpening filters with transform-based fusion, the researchers sought to improve edge preservation, image sharpness and the representation of high-frequency information while maintaining computational efficiency. The supplied abstract does not report processing time, hardware specifications, memory consumption or a direct runtime comparison, so the efficiency claim cannot be independently assessed from the accessible material.

Among the evaluated fusion techniques, “DCT + HB” delivered the maximum reported performance across the study’s statistical measures. It combined the highest PSNR with the lowest RMSE, along with a mean absolute error of 1.6306 and entropy of 0.9324. The abstract says the outcome was supported by both qualitative and quantitative comparisons with existing fusion approaches. These results show that the DCT-plus-high-boost configuration performed best within the reported benchmark, rather than establishing universal superiority over every image-fusion method. The evaluation also provides evidence that the hybrid pipeline can improve the assessed features of fused grayscale outputs, but the accessible report does not supply a complete account of the statistical testing or uncertainty surrounding the numerical comparisons.

MFSIF also represents a different technical direction from the deep-learning fusion systems discussed in the source material. Recent approaches based on convolutional neural networks and generative adversarial networks have been developed to provide improved visual quality and feature extraction, whereas the proposed framework uses a non-neural pipeline of sharpening filters and wavelet- and cosine-transform operations. The study’s central reported result is that sharpening the source images before transform-based fusion can complement the subsequent combination step, producing more clearly defined edges and stronger representation of fine detail. That claim is supported by the four-dataset experiment, not by validation in medical diagnosis, surveillance analysis, remote sensing or other deployed systems. Whether the method performs similarly with color imagery, additional sensors, different resolutions or real-world imaging conditions remains outside the scope of the reported evaluation.

Subject of Research: Evaluation of a hybrid filtering and transform-domain framework for grayscale multi-focus image fusion.

Article Title: Multi-Focus Sharpened Image Fusion (MFSIF): a novel framework based on high boost filtering and discrete cosine transformation approaches

Article References: Shah, M., Khan, S. S., Danyal, M. M., Khan, M., Shah, M. A., & Arshad, M. (2026). Multi-Focus Sharpened Image Fusion (MFSIF): a novel framework based on high boost filtering and discrete cosine transformation approaches. Multimedia Tools and Applications, 85(11), Article 819. https://doi.org/10.1007/s11042-026-21979-z

Image Credits: AI Generated

DOI: 10.1007/s11042-026-21979-z

Keywords: multi-focus image fusion, high-boost filtering, unsharp masking, discrete cosine transform, discrete wavelet transform, image sharpening, edge preservation, grayscale imaging, PSNR, signal processing, Multi-Focus, Sharpened