PCNN Model Guided by Saliency Mechanism for Image Fusion in Transform Domain.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Heterogeneous image fusion challenges exist between time-of-flight (ToF) and visible light (VIS) images in orchard environments.
• Existing pulse coupled neural network (PCNN) models have limitations, including manual parameter settings, lack of adaptive termination, and susceptibility to image fluctuations, leading to artifacts like blurring and unclear edges.

Purpose of the Study:

• To propose an advanced image fusion method for ToF and VIS images that addresses the shortcomings of current PCNN models.
• To enhance the fusion quality for binocular acquisition systems in complex natural scenes, particularly orchards.

Main Methods:

• A saliency-guided PCNN transform domain method is proposed.
• Non-subsampled shearlet transform decomposes registered images.
• ToF low-frequency components are simplified using PCNN-based multi-lighting segmentation.
• A significance function based on first-order Markov mutual information defines the termination condition.
• A momentum-driven multi-objective artificial bee colony algorithm optimizes PCNN parameters.
• Low-frequency components are fused using a weighted average rule, and high-frequency components using improved bilateral filters.

Main Results:

• The proposed algorithm demonstrates superior fusion performance on ToF confidence and VIS images compared to existing methods.
• Objective evaluation using nine indicators confirms the algorithm's effectiveness in natural scenes.
• The method successfully fuses heterogeneous images from complex orchard environments.

Conclusions:

• The developed saliency-guided PCNN transform domain fusion method effectively overcomes the limitations of traditional PCNN models.
• The algorithm provides high-quality fusion results suitable for heterogeneous image fusion in challenging natural landscapes like orchards.