Neural Network Image Section Identification via Back-Propagation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image processing methods, such as Generative Adversarial Networks (GAN), face limitations in enhancing resolution of specific image sections efficiently, leading to resource wastage and quality issues, as they often enhance the entire image instead of focusing on targeted areas, and fail to handle arbitrary shaped portions without considering neighboring pixels.
Innovation Solution
A system and method that utilize a neural network to identify and modify target image sections within images by extracting user-defined attributes, comparing them with a pattern attribute table, and applying back-propagation techniques to selectively enhance specific features of the image sections based on user input, using a processor and memory to execute instructions for image processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional GAN enhances resolution of the whole image, then the resolution of the entire image is improved, but resource consumption and processing time increase unnecessarily
Solution Approach 1:
The patent divides the image into multiple sections and identifies only the relevant sections that require resolution enhancement. By segmenting the image processing task and applying enhancement only to identified sections rather than the entire image, the system reduces computational resources and processing time while maintaining the resolution improvement benefit for the areas that matter most to the user.
2Measurement precision
If conventional methods enhance resolution of full image, then overall image quality is improved, but memory requirements increase significantly
Solution Approach 1:
The patent extracts and identifies only the relevant sections from the full image that require resolution enhancement. By taking out only the necessary portions for processing rather than handling the complete image, the system reduces memory requirements while still achieving the desired resolution improvement for the extracted sections.
3Quantity of substance
If conventional methods process down sampled images, then more training data is obtained, but relative relevance features are lost
Solution Approach 1:
The patent applies different processing qualities to different parts of the image by identifying relevant sections and applying resolution enhancement specifically to those areas. This local quality approach ensures that important features maintain their relevance and detail while reducing unnecessary processing of irrelevant areas, thereby preserving feature importance without requiring excessive down-sampling.
Data Source
AI summary
A system and method for identifying and subsequently modifying target sections within images is disclosed. The method includes receiving a user request that includes a target image and a user input, such that the user input includes an action corresponding to the target image. The method further includes extracting a plurality of target image attributes from the user input and comparing the plurality of target image attributes with a set of attributes within a pattern attribute table. The pattern attribute table includes mapping of each of the set of attributes to an associated set of activated neurons within a neural network. Comparing the plurality of target image further includes identifying a set of activated neurons mapped to the matching attribute as neuron activations corresponding to the plurality of target image attributes. The method includes identifying a target image section based on the set of activated neurons using a back-propagation technique.