Research on Image Segmentation Algorithm Based on Multimodal Hierarchical Attention Mechanism and Genetic Neural...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Multimodal tasks leveraging attention mechanisms and language processing encounter significant challenges.
• Existing image segmentation models struggle with static and dynamic feature maps in image sequences.

Purpose of the Study:

• To apply an image segmentation algorithm for data completion and 3D scene reconstruction.
• To address limitations in attention enhancement feature extraction modules regarding module count and parameter size.

Main Methods:

• Utilized a multimodal hierarchical attention mechanism combined with a genetic neural network.
• Implemented a recursive mechanism for feature extraction to manage model parameters efficiently.
• Calculated pixel differences in color space against target image averages.

Main Results:

• Achieved an accuracy of 89.7% for the improved image saliency detection algorithm.
• Reduced the average single-point pixel difference to the target image to 0.132.
• Demonstrated enhanced practicability and reliability of the image segmentation model.

Conclusions:

• The proposed multimodal hierarchical attention and genetic neural network approach effectively enhances image segmentation for data completion and 3D reconstruction.
• The recursive feature extraction mechanism successfully mitigates issues with large parameter counts and module numbers.
• The improved algorithm offers a more practical and reliable solution for complex image-based tasks.