Vector Line Drawing Network for Rough Raster Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vectorization software require clean line drawings on a white background, making it tedious to convert rough raster images directly into vector line drawings, often necessitating a two-step process of cleaning and then vectorizing, which is inefficient and difficult to automate.
Innovation Solution
A machine learning-based model using a recurrent neural network (RNN) trained end-to-end to convert rough raster images directly into vector line drawings, generating strokes stroke by stroke with a dynamic drawing window that moves and scales across the image, allowing for direct conversion without prior cleaning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If existing vectorization software is used to convert raster images to vector line drawings, then the conversion can be performed, but the input raster images must be clean line drawings on a white background, requiring a two-step process of cleaning and then vectorizing
Solution Approach 1:
The patent combines the image cleaning function and vectorization function into a single integrated system. The neural network simultaneously performs denoising, background removal, and vector line drawing generation in one end-to-end process, eliminating the need for separate cleaning and vectorizing steps.
Solution Approach 2:
The vectorization system is designed to handle multiple types of input images (clean line drawings, rough sketches, photographs) and performs multiple functions (cleaning, vectorization) within a single model, making it universally applicable without requiring preprocessing for different input types.
2Manufacturing precision
If a two-step process of cleaning and then vectorizing is used, then accurate vector line drawings can be obtained, but the process becomes tedious and difficult to automate
Solution Approach 1:
The system performs preliminary cleaning actions automatically as part of the vectorization process. The neural network inherently handles denoising and background removal during training and inference, so no separate preliminary cleaning step is needed by the user.
Solution Approach 2:
The vectorization system is self-sufficient and automatically handles all preprocessing and conversion tasks without requiring user intervention for image cleaning. The model autonomously processes rough sketches and photographs directly into vector line drawings.
3Productivity
If rough raster images are converted directly to vector line drawings without cleaning, then computational efficiency is improved, but the accuracy and quality of the vector output may deteriorate
Solution Approach 1:
The patent replaces traditional mechanical image processing methods (filtering, thresholding, manual cleaning) with a neural network-based system. The deep learning model automatically learns to clean and vectorize images through end-to-end training, achieving both efficiency and quality without manual preprocessing.
Solution Approach 2:
The system changes the approach from sequential processing (cleaning then vectorizing) to simultaneous processing through end-to-end training. The loss function combines both cleaning accuracy and vectorization quality metrics, allowing the model to optimize both objectives together during training.
Data Source
AI summary
The present disclosure describes methods and devices for generating a vector line drawing. A vector line drawing network may include a machine learning-based model that is trained to convert a raster image to a vector line drawing directly. The vector line drawing network may be trained end-to-end, using supervised learning, where only raster images are used as training data. A vector line drawing is generated stroke by stroke, over a series of time steps. In each time step, a dynamic drawing window is moved and scaled across the input raster image to sample a patch of the raster image, and a drawing stroke is predicted to draw a stroke in a corresponding patch in the canvas for the vector line drawing. The image patches are pasted in the canvas to assemble a final vector line drawing that corresponds to the input raster image.