Cleanup Sketched Drawings: Deep Learning-Based Model.

Area of Science:

• Computer Vision
• Digital Image Processing
• Machine Learning

Background:

• Rough sketches are fundamental artistic tools but often contain impurities and low resolution.
• Enhancing and simplifying these sketches is crucial for clear visual representation.

Purpose of the Study:

• To propose a deep learning model for automatic simplification and enhancement of rough raster drawings.
• To address issues like noise, unwanted background, and low resolution in sketches.

Main Methods:

• Utilized a fully convolutional network (FCNN) model for sketch simplification.
• Trained the FCNN model on three publicly available raster image datasets.
• Employed the mean squared error (MSE) metric for performance evaluation.

Main Results:

• The FCNN model successfully generates high-quality simplified sketch images from any input size.
• The model effectively removes noise and unwanted background, improving curve simplification.
• An enhanced FCNN model demonstrated improved accuracy, reducing prediction error by 0.08% compared to existing methods.

Conclusions:

• Deep learning, specifically FCNNs, offers an efficient and effective automated solution for cleaning and improving rough sketches.
• The proposed method significantly enhances the quality and usability of digital drawings.
• The study validates the potential of AI in artistic image processing and refinement.