A method for improving the image stitching accuracy of a scanning and writing pen

By constructing a convolutional neural network with a backbone, neck, and head structure, and combining it with an FPN network for image stitching, the problems of low image stitching efficiency and insufficient feature capture in existing technologies are solved, and high-precision image stitching results are achieved.

CN119863356BActive Publication Date: 2026-06-05INGENIC SEMICON CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
INGENIC SEMICON CO LTD
Filing Date
2023-10-19
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

Existing image stitching methods are inefficient and error-prone when stitching large and high-resolution images. Deep learning-based methods cannot capture features at different scales well during feature upsampling and require complex post-processing.

Method used

A simple convolutional neural network is used, employing a Backbone, Neck, and Head structure, combined with an FPN network for feature extraction and upsampling. The model is trained using a cross-entropy loss function, and keypoint indices are directly output during the inference stage, avoiding post-processing of non-maximum suppression (NMS).

Benefits of technology

It improves image stitching accuracy, enhances model performance and stability, simplifies the processing flow, and reduces reliance on parameter adjustments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119863356B_ABST
    Figure CN119863356B_ABST
Patent Text Reader

Abstract

The application provides a method for improving the image splicing accuracy of a scanning and writing pen, comprising the following steps: S1, constructing a convolutional neural network; S2, training a neural network model, that is, training the convolutional neural network constructed in the step S1; and S3, an inference stage: processing the network output result. The method uses a simple convolutional model, and the accuracy is greatly improved under the premise of ensuring the speed. Compared with the traditional image splicing method ORB, the accuracy is greatly improved, compared with the method based on deep learning SuperPoint, the method is simple to train, solves the multi-scale feature problem, and obtains the result in an end-to-end mode, and does not need complex post-processing.
Need to check novelty before this filing date? Find Prior Art