Adaptive Image Filtering Using Segmented Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in applying filters to input image data based on image characteristics without increasing network size, leading to inefficiencies in coding and decoding processes.
Innovation Solution
An image filtering apparatus utilizing multiple dedicated neural networks and a common neural network, where dedicated networks act on input data based on filter parameters and the common network processes output data, allowing for adaptive filtering according to image characteristics while maintaining a reduced network size.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a filter is applied in accordance with characteristics of the input image data using a single neural network, then the network can process various image types, but the network size increases
Solution Approach 1:
The patent divides the filtering system into multiple dedicated neural networks, each specialized for specific image characteristics (e.g., one network for luminance data, another for chrominance data). This segmentation allows each network to be smaller and more specialized, reducing overall network size while maintaining adaptability through selective network application based on input image characteristics.
Solution Approach 2:
The patent applies different filtering approaches to different regions or types of image data by using multiple dedicated networks. Each network is optimized for specific local characteristics (such as different quantization parameter ranges or image regions), enabling adaptability without requiring a single large universal network.
2Manufacturing precision
If multiple dedicated neural networks are used to process different image characteristics, then filtering accuracy improves, but device complexity increases
Solution Approach 1:
The patent introduces a common neural network that can be shared across multiple dedicated networks. This common network performs universal processing functions that are needed across different image types, reducing overall system complexity while maintaining the benefits of multiple specialized networks. The common network serves multiple purposes, handling tasks that are common to different image characteristics.
Solution Approach 2:
The patent combines multiple dedicated neural networks with a common neural network into an integrated filtering system. The dedicated networks handle specific image characteristics while the common network provides shared processing capabilities, merging the advantages of specialization with the efficiency of shared resources, thereby managing complexity while maintaining accuracy.
Data Source
AI summary
To apply a filter to input image data in accordance with an image characteristic. A CNN filter includes a neural network configured to receive an input of one or multiple first type input image data and one or multiple second type input image data, and output one or multiple first type output image data, the one or multiple first type input image data each having a pixel value of a luminance or chrominance, the one or multiple second type input image data each having a pixel value of a value corresponding to a reference parameter for generating a prediction image and a differential image, the one or multiple first type output image data each having a pixel value of a luminance or chrominance.