图像处理超参数优化方法、系统、设备及存储介质
By training image processing models for both the debugging and comparison devices and optimizing hyperparameter combinations, the problem of perspective differences between the debugging and comparison devices was solved, thus improving the effectiveness and efficiency of image processing.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SUZHOU KEYUAN SOFTWARE TECH DEV
- Filing Date
- 2023-03-23
- Publication Date
- 2026-07-17
AI Technical Summary
In existing technologies, the difference in perspective between debugging and comparison devices leads to poor hyperparameter optimization results in image processing, and traditional methods are time-consuming and difficult to guarantee global or local optimality of parameters.
By acquiring training image data from debugging and comparison devices respectively, first and second image processing models are trained. These models are then used to simulate the image processing algorithms of the debugging and comparison devices, optimize the hyperparameter combination, and use the uformer model and EnhanceNet model for feature extraction and image correction to reduce the effects of viewpoint differences and lens distortion.
This technology simulates the image processing style of a comparative device when optimizing hyperparameter combinations, improving the hyperparameter optimization effect, reducing the impact of perspective differences and lens distortion, and enhancing the quality and efficiency of image processing.
Smart Images
Figure CN116347251B_ABST