Deep Learning Image Enlarging Circuit with Super-Resolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image enlarging technologies fail to enhance the resolution of enlarged images, resulting in blurriness, unclear edges, and noise, especially as digital resolution improves from HD to ultra HD, necessitating a mechanism to adjust and enhance images based on local characteristics.
Innovation Solution
An image enlarging apparatus and method incorporating a deep learning mechanism with a deep learning circuit, image concatenating circuit, and super-resolution image enlarging circuit, which downsizes, analyzes, and reallocates weights for image upsizing, concatenating the input image with an adjusting map to perform super-resolution enlargement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Volume of moving object
If conventional image enlarging technologies are used, then the image size is increased, but the resolution deteriorates resulting in blurriness and noise
Solution Approach 1:
The patent applies preliminary action by performing image downsizing before the main enlarging process. The input image is downsized to a smaller dimension, processed through deep learning networks to extract features and generate weighting maps, then upsized back to the target resolution. This preliminary downsizing- processing- upsizing sequence enables the system to capture local image characteristics more effectively and apply adaptive weighting, thereby achieving high-resolution output without the blurriness and noise that plague conventional direct enlarging methods.
2Manufacturing precision
If deep learning processing is applied to the entire image, then local characteristics are enhanced, but the processing complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the image processing into distinct stages and components. The deep learning network is segmented into multiple functional blocks: a downsizing module that reduces image dimension, characteristic analysis modules that extract different local features (edges, textures, colors), weighting map generation modules, and an upsizing module. Each segment processes specific aspects of the image independently, allowing the system to enhance local characteristics effectively while managing computational complexity through modular architecture.
Solution Approach 2:
The patent implements local quality by applying different processing strategies to different regions of the image based on local characteristics. The deep learning network generates separate weighting maps for different image features (e.g., edge regions, texture regions, smooth regions), and these weighting maps are applied locally to enhance specific areas appropriately. This allows the system to preserve important local details while avoiding over-processing in regions that don't require enhancement, thereby improving local image quality without uniformly increasing overall processing complexity.
Data Source
AI summary
The present disclosure discloses an image enlarging apparatus having deep learning mechanism. A deep learning circuit includes an image downsizing circuit, an image characteristic analyzing circuit, a weighting reallocating circuit and an image upsizing circuit. The image downsizing circuit downsizes an input image to generate a downsized image. The image characteristic analyzing circuit analyzes the downsized image according to image characteristics to generate categorized images. The weighting reallocating circuit performs weighting reallocating on the categorized images according to image weighting parameters corresponding to the image characteristics to generate weighting reallocated images. The image upsizing circuit upsizes the weighting reallocated images to generate adjusted images. A concatenating circuit concatenates the input image and the adjusted images to generate concatenated images. A super-resolution enlarging circuit performs super-resolution enlarging on the concatenated images to generate an output image.