Image Processing Masking for Accurate Neural Target Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural networks struggle to accurately recognize pretrained target areas when input data is not normally distributed, often due to masking processes that alter color values, leading to inaccurate training and detection.
Innovation Solution
Separate detection of pretrained target areas and regions of interest, using unmasked image data to ensure normal distribution, allowing for improved training and hardware performance, and enabling parallel processing of multiple regions of interest.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If an image mask is applied to remove non-interest image parts before inputting to the neural network, then the processing focuses on relevant regions, but the color values become non-normally distributed leading to inaccurate target area detection
Solution Approach 1:
The patent segments the image processing into two independent detection tasks: target area detection using unmasked image data, and region of interest detection using the same unmasked data. This segmentation allows each task to operate on normally distributed data without the distorting effect of masking, resolving the contradiction between processing efficiency and detection accuracy.
Solution Approach 2:
Instead of the conventional approach of masking first then detecting, the patent inverts the sequence by detecting target areas on unmasked data first, then determining regions of interest. This inversion preserves the normal distribution of color values for accurate detection while still enabling focused processing on relevant regions.
2Device complexity
If the same masked image data is used for both target area detection and region of interest detection, then the processing is simplified, but the target area detection accuracy deteriorates due to non-normal color value distribution
Solution Approach 1:
The patent divides the detection process into two separate detection operations that both operate on unmasked image data. The first detection identifies target areas, and the second detection identifies regions of interest. This segmentation maintains data integrity for accurate target detection while still achieving processing efficiency through separate optimized operations.
Solution Approach 2:
The patent uses the same unmasked image data as input for both detection operations, effectively creating parallel processing paths from the same data source. This copying approach ensures both detections operate on normally distributed data without the need to create and manage multiple masked versions of the image.
3Ease of manufacture
If masking is applied to remove non-interest regions, then training can focus on relevant areas, but the training accuracy deteriorates because the input data is no longer normally distributed
Solution Approach 1:
The patent segments the training process into two independent training operations: one for target area detection and another for region of interest detection. Both operations are trained on unmasked image data, preserving the normal distribution of color values and ensuring high training accuracy while maintaining simplicity through focused training objectives.
Solution Approach 2:
The patent performs target area detection on unmasked data before applying any masking operations. This preliminary detection ensures that the training data maintains normal distribution characteristics, allowing the neural network to learn accurate target area features without the distorting effects of masking during the training phase.
Data Source
AI summary
The invention relates to a method for processing at least one image (1), wherein the method comprises the following steps: receiving of image data. inputting the unmasked image data to an artificial neural network for detecting one or more pretrained target areas (3). detecting one or more region of interests (2) in the image (1) and creating an image mask that is dependent on the one or more region of interests (2) and applying the created image mask on the image data for removing an image region (4) that does not comprise the one or more region of interests (2).