Distorted Image Generation for Deep Learning Robustness
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning models struggle with accurately identifying and segmenting images that are distorted, as the degree and type of distortion can vary, leading to inaccurate recognition and segmentation.
Innovation Solution
A method for generating distorted images by calculating new pixel coordinates and values based on distortion coefficients, allowing for the creation of images with varying degrees and types of distortion, which can be used to train deep learning models to improve their robustness and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a deep learning model is trained on undistorted images only, then the model structure can be kept simple, but the recognition accuracy deteriorates when processing distorted images
Solution Approach 1:
The patent applies preliminary action by pre-processing images to generate distorted versions before training the model. The system creates multiple distorted versions of each image using different distortion coefficients and types, then trains the model on this augmented dataset. This prepares the model in advance for various distortion scenarios, enabling it to maintain high recognition accuracy when processing distorted images during deployment.
2Measurement precision
If camera distortion correction is applied, then image accuracy for recognition can be improved, but the system becomes more dependent on correction algorithms and loses robustness
Solution Approach 1:
The patent converts the harmful effect of image distortion into a beneficial training resource. Instead of treating distortion as a problem to be corrected, the system generates distorted images using various distortion coefficients and types, then uses these distorted images as training data. This transforms the previously harmful distortion into a useful element that enhances model robustness and enables accurate recognition of distorted images without requiring pre-correction.
3Measurement precision
If distortion correction is performed before recognition, then recognition accuracy can be maintained, but the processing time and system complexity increase
Solution Approach 1:
The patent performs distortion handling in advance during the training phase rather than during real-time processing. By pre-generating distorted images with various coefficients and training the model on this augmented dataset, the system eliminates the need for runtime distortion correction. This preliminary action transfers the computational burden to the offline training stage, enabling fast, accurate recognition during deployment without additional processing time.
Data Source
AI summary
A method for generating distorted images is applied in an electronic device, obtains first pixel coordinates of undistorted images and a first pixel value of the first pixel coordinates, and selects an arbitrary distortion center coordinate. The distance between the coordinate of the distortion center and each first pixel coordinate is calculated, and second pixel coordinates corresponding to the first pixel coordinates are calculated according to distortion coefficient, the first pixel coordinates, and the distance. The first pixel value of each first pixel coordinates is taken as the second pixel value of each second pixel coordinates, and distorted images from undistorted images are generated for training purposes according to the second pixel coordinates and the second pixel values.