Panoramic Image Enhancement Model Training Data Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Panoramic images captured by panoramic cameras suffer from poor sharpness, resolution, and hue differences compared to planar images from high-quality cameras like digital SLR cameras.
Innovation Solution
An image processing method that utilizes training data sets containing low-quality and high-quality images with the same shooting content to construct an image enhancement model, which improves the quality of panoramic images by minimizing the difference between the output and the high-quality images, using convolutional neural networks or generative adversarial networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If panoramic camera is used to capture wide field of view, then field of view is increased, but image quality (sharpness, resolution, hue) deteriorates
Solution Approach 1:
The patent introduces an image enhancement model as an intermediary between the low-quality panoramic image and the desired high-quality output. This neural network model acts as a mediator that transforms the degraded panoramic image into an enhanced version with improved sharpness, resolution, and hue accuracy, effectively decoupling the trade-off between wide field of view and image quality
Solution Approach 2:
The patent changes the parameters of the panoramic image through the image enhancement model, which adjusts key image quality parameters including sharpness (edge definition), resolution (pixel detail), and hue (color accuracy). By applying learned transformations from training data, the model modifies these parameters to achieve high-quality output while maintaining the wide field of view capability
2Manufacturing precision
If image enhancement model is trained with multiple datasets, then image quality improvement is enhanced, but training time and computational resources increase
Solution Approach 1:
The patent performs preliminary actions by pre-training the image enhancement model using multiple datasets containing paired low-quality and high-quality images. This offline training phase prepares the model in advance with learned enhancement patterns, so that during actual deployment, the model can quickly process images without requiring extensive real-time computation, thus reducing operational time loss
Solution Approach 2:
The patent merges multiple training datasets into a unified training process, combining diverse image pairs (panoramic-planar, different resolutions, various lighting conditions) to create a comprehensive training corpus. This merging approach allows the model to learn robust enhancement patterns across different scenarios simultaneously, improving overall image quality capability while efficiently utilizing training resources
Data Source
AI summary
Disclosed is an image processing method including steps of obtaining a plurality of sets of training data, each set of training data containing data of first and second images, image quality of the second image being higher than that of the first image, shooting contents of the first image being the same model, letting the data of the first image be an input of the model, and utilizing the plurality of sets of training data to train the model until a difference feature value between an output of the model and the data of the second image is minimum; and inputting data of a third image to be processed into the model so as to output data of a fourth image after image enhancement, image quality of the fourth image being higher than that of the third image.