Image Upscaling via Fixed Block Division for Partial Regions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image upscaling techniques face challenges in efficiently processing large images, particularly in ML-based upscaling, where memory and computation requirements increase exponentially with image size. Additionally, methods that divide images into blocks can result in varying processing accuracy and time depending on the block setting method.
Innovation Solution
An image processing apparatus that divides an image into blocks and performs upscaling processing on each block, combining the upscaled blocks into a single image. In generating an upscaled image of a partial region, the apparatus either divides the partial region into blocks at predetermined positions regardless of its position in the original image or sets a processing region that contains the partial region and performs upscaling on blocks within this region.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If image upscaling is performed by dividing into blocks, then memory and computation requirements are reduced, but processing accuracy and processing time vary depending on block setting method
Solution Approach 1:
The image is divided into multiple blocks for parallel processing, reducing memory and computation requirements. Each block is processed independently through the neural network, enabling scalable processing of large images without requiring excessive resources.
Solution Approach 2:
Different block size configurations are applied based on the specific processing requirements and image characteristics. The system can adjust block dimensions and overlapping regions locally to optimize both processing efficiency and accuracy for different regions of the image.
2Length of stationary object
If image upscaling is performed by dividing into blocks, then processing of large images becomes feasible, but processing time varies depending on block setting method
Solution Approach 1:
The large image is segmented into multiple smaller blocks that can be processed in parallel, making it feasible to upscale images of various sizes including very large images. The parallel processing of blocks significantly reduces overall processing time compared to sequential processing of the entire image.
Solution Approach 2:
The block configuration (size, number of blocks, overlapping regions) can be dynamically adjusted based on the image size, available computational resources, and desired processing speed. This allows optimization of processing time for different scenarios while maintaining feasibility for large images.
3Productivity
If blocks are set according to partial region position, then processing may be optimized for that region, but processing accuracy and results become inconsistent across different positions
Solution Approach 1:
A universal block processing approach is applied where the same neural network model and processing pipeline are used for all blocks regardless of their position in the image. This ensures consistent processing results across different regions while maintaining efficiency through standardized operations.
Solution Approach 2:
The system can adjust processing parameters such as block size and overlapping region proportions based on the partial region's characteristics and position. By dynamically changing these parameters, the system maintains processing consistency across different image regions while optimizing for the specific requirements of each region.
Data Source
AI summary
An image processing apparatus that upscales a partial region of an image using an upscaling method in which upscaling is performed in a block unit is disclosed. The apparatus divides an image into blocks, performs upscaling processing on each block, and then combines upscaled images of the blocks into one upscaled image. In a case where an upscaled image of a partial region of an original image is to be generated, the apparatus divides the partial region into a plurality of blocks at predetermined positions regardless of a position of the partial region in the original image.