Application of Digital Image Based on Machine Learning in Media Art Design

Area of Science:

• Digital Media Art
• Computer Vision
• Image Processing

Background:

• Digital image technology revolutionizes traditional media art expression.
• Expressive force in digital media art relies on effective image manipulation.
• Existing methods lack efficient subpixel sample generation for media art.

Purpose of the Study:

• To propose a partial-pixel interpolation technique for digital media art.
• To develop a subpixel sample generation algorithm for media art images.
• To enhance interframe prediction accuracy in media art through motion compensation.

Main Methods:

• A partial-pixel interpolation technique based on convolutional neural networks (CNNs) is proposed.
• A subpixel sample generation algorithm using Gaussian low-pass filtering and polyphase sampling is developed.
• Pixel motion compensation is framed as an interframe regression problem for bidirectional prediction.

Main Results:

• The proposed CNN-based technique effectively performs partial-pixel interpolation for media art.
• The subpixel sample generation algorithm addresses the challenge of unobtainable subpixel data.
• The generalized partial-pixel interpolation model improves bidirectional prediction accuracy.

Conclusions:

• The developed algorithm offers a flexible and efficient approach to applying trained digital images in media art design.
• This research advances digital image processing techniques for enhanced media art expression.
• The study bridges the gap between advanced image processing and creative media art applications.