.png){kind=icon}
.png){kind=icon}
Return
Depth from focus using directional spherical difference filter and vector to scalar fusion
.png){kind=icon}
PRE
AI
DOI：10.1016/j.jvcir.2026.104794.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Traditional Depth from Focus (DFF) methods apply a focus measure (FM) to a focus-varying image sequence to generate a focus volume (FV), which encodes per-pixel focus information. Typically, a single FM is used to obtain a scalar focus response per pixel; however, relying on a single response is vulnerable to noise and disruptions. Some approaches extract multiple responses per pixel, yielding a vector of focus responses and generating multiple FVs, but these methods often combine the FVs using simple operations like summing, median filtering, or selecting the maximum response, which loses valuable focus information. To address these challenges, we propose a novel framework that first generates multiple FVs by employing our proposed Directional Spherical Difference Filter (DSDF) to obtain vector focus responses, and then integrates these responses into a single scalar value using a novel Vector to Scalar Fusion (VSF) technique. In the VSF process, difference vectors are computed by subtracting a central vector from those within a local window, and the spread of these vectors is used as a scaling factor for the resultant magnitude, producing an enhanced scalar focus value per pixel that leads to a more accurate depth map. Evaluation on five diverse datasets, comprising more than 1200 scenes and 17,000 images, demonstrates our method's effectiveness, outperforming 19 well-established methods in both quantitative and qualitative comparisons.
Keywords:
Depth from focus
Focus measure
VeCtor to Scalar Fusion
Journal
IF:
3.1
Papers:
414
Citations:
5.6K