Image Fusion Using Depth Convolutional Neural Networks for Night Sky Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Image segmentation and fusion techniques yield poor results when applied to night skies or skies with dense fog, failing to effectively enhance image information or add content.
Innovation Solution
An image processing method using a depth convolutional neural network for segmentation, followed by binary mask image creation and image fusion, incorporating Gaussian blurring and color correction to improve image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If image segmentation and fusion techniques are performed on a sky in the night or a sky when there is dense fog, then the image processing is completed, but the image acquired through fusion has a poor effect
Solution Approach 1:
The patent changes the parameter representation from raw pixel values to depth maps and semantic feature layers. By transforming the image data into depth space and multi-scale feature representations, the system can effectively process challenging conditions like night skies and dense fog that would otherwise degrade traditional image fusion performance.
Solution Approach 2:
The patent introduces a new dimension by incorporating depth information through depth convolutional neural networks. Instead of working solely in the 2D image plane, the system processes images in 3D depth space and combines multiple feature layers, enabling effective segmentation and fusion where traditional 2D methods fail under adverse conditions.
2Measurement precision
If depth convolutional neural network is used for segmentation, then segmentation accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent applies segmentation at multiple levels: first segmenting the image into semantic feature layers at different scales, then processing each layer through the depth convolutional neural network. This hierarchical segmentation approach improves accuracy while managing computational complexity by dividing the processing into manageable stages.
Solution Approach 2:
The system extracts semantic feature layers at two or more scales, processing only the necessary portions of the image at each scale. By selectively processing feature layers rather than the entire image at full resolution, the system achieves high segmentation accuracy while controlling computational resource consumption.
Data Source
AI summary
An image processing method, an image processing apparatus, and a computer storage medium are disclosed. The image processing method includes: acquiring a first still image; acquiring a first moving image comprising a plurality of image frames; performing segmentation on the first still image to obtain a first feature region in the first still image; acquiring a binary mask image of the first feature region; and performing image fusion on the first still image and the plurality of image frames based on the binary mask image to obtain a second moving image.