Image Recognition Model Generation via Region Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In image recognition, particularly in face recognition, identifying appropriate feature data is crucial for pre-training and fine-tuning image recognition models. However, when appropriate feature data cannot be identified, the model fails to yield a fine result, especially when face information is incomplete.
Innovation Solution
A method is developed to generate an image recognition model by obtaining a source image, cutting out a first image from a specific region, and pre-training a preliminary image recognition model using the cut source image and associated label data. The model is then fine-tuned to generate the image recognition model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If standard transfer learning process is used with pre-training and fine-tuning, then model training can be performed, but the model fails to yield fine results when appropriate feature data cannot be identified or face information is incomplete
Solution Approach 1:
The patent segments the source image into multiple regions including a first region containing the first image and other regions. By cutting out the first image from the first region to generate a cut source image, the method focuses on specific facial features while excluding interfering elements, thereby improving recognition accuracy for incomplete face information
Solution Approach 2:
The patent extracts the first image from the source image by cutting out the first region. This extraction process isolates the essential facial features needed for training, removing background and other non-essential elements that could interfere with model learning, thus enhancing the model's ability to handle incomplete face data
2Quantity of substance
If complete source images are used for pre-training, then sufficient data is available for training, but the model cannot effectively handle noisy or incomplete input images
Solution Approach 1:
The patent applies local quality by generating different processed images from different regions of the source image. The first processed image is generated from the first region containing the first image, while second processed images are generated from other regions. This allows the model to learn from diverse local features while maintaining high quality feature data for training
Solution Approach 2:
The patent converts potentially harmful incomplete or noisy face information into beneficial training data by using image processing techniques. By cutting out specific regions and generating processed images, the method transforms incomplete information into focused feature data that enhances model robustness against noise and incompleteness
Data Source
AI summary
The invention provides a method of generating an image recognition model and an electronic device using the method. The method includes the following. A source image is obtained; a first image is cut out of a first region of the source image to generate a cut source image; a preliminary image recognition model is pre-trained according to feature data and label data, in which the feature data is associated with the cut source image, and the label data is associated with the first image; and the pre-trained preliminary image recognition model is fine-tuned to generate the image recognition model. The method of generating the image recognition model and the electronic device provided by the invention may correctly restore an input image.