Deformed Image Recognition via Spatial Transformation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object recognition algorithms based on neural networks fail to accurately recognize objects in deformed images due to deformations such as tilt, scaling, and perspective transformation, leading to low recognition accuracy.
Innovation Solution
An object recognition method and apparatus that involves inputting an image into a preset localization network to obtain localization parameters, performing spatial transformation to correct the image, and then inputting the corrected image into a recognition network for classification, using convolutional layers and fully connected layers to extract and regress features, and apply affine, perspective, or thin plate spline transformation algorithms as needed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing object recognition algorithms directly classify objects in deformed images, then the recognition process is simple and fast, but the accuracy of object recognition is low
Solution Approach 1:
The patent applies preliminary action by performing spatial transformation to correct image deformation before object recognition. The system first detects deformation parameters (tilt angle, scaling factor, perspective transformation) and applies correction transformations to obtain a corrected image, then performs recognition on the corrected image. This preliminary correction step eliminates the negative impact of deformation on recognition accuracy while maintaining a relatively simple overall framework.
2Measurement precision
If spatial transformation is applied to correct image deformation, then object recognition accuracy is improved, but the processing time and computational complexity increase
Solution Approach 1:
The patent segments the image processing task into distinct stages: deformation detection (obtaining tilt angle, scaling factor, perspective parameters), spatial transformation correction, and object recognition. By separating the correction step from recognition and using targeted transformation algorithms based on detected deformation types, the system minimizes unnecessary computational overhead while ensuring accurate correction.
Solution Approach 2:
The system dynamically adjusts transformation parameters based on detected deformation characteristics. Instead of applying fixed or exhaustive transformation sequences, the patent calculates specific transformation parameters (rotation angle, scaling ratio, perspective matrix) from detected deformation features and applies only the necessary corrections, reducing computational time while maintaining correction accuracy.
Data Source
AI summary
An object recognition method and apparatus for a deformed image are provided. The method includes: inputting an image into a preset localization network to obtain a plurality of localization parameters for the image, wherein the preset localization network comprises a preset number of convolutional layers, and wherein the plurality of localization parameters are obtained by regressing image features in a feature map that is generated from a convolution operation on the image; performing a spatial transformation on the image based on the plurality of localization parameters to obtain a corrected image; and inputting the corrected image into a preset recognition network to obtain an object classification result for the image. In the process of the neural network based object recognition, the embodiment of the present application first transforms the deformed image that has deformation, and then performs the object recognition on the transformed image.