Deep-learning-based pixel compensation algorithm for local dimming liquid crystal displays of quantum-dot backlights.

Area of Science:

• Display Technology
• Image Processing
• Artificial Intelligence

Background:

• Local dimming is crucial for high contrast and low power in liquid crystal displays.
• Current methods require significant processing power and backlight luminance data for pixel compensation.
• Adjacent backlight blocks complicate existing compensation algorithms.

Purpose of the Study:

• To develop a deep-learning-based local dimming algorithm that bypasses the need for backlight dimming level information.
• To generate a compensated image directly from an input image for improved efficiency.
• To reduce computational complexity in local dimming techniques.

Main Methods:

• A U-net based convolutional neural network architecture was employed to maintain high-resolution features.
• Skip-connections were utilized in the up-sampling paths of the network.
• Bi-linear interpolation was incorporated to reduce model parameters without compromising image quality.

Main Results:

• The proposed deep-learning algorithm successfully generates compensated images directly from input images.
• The network architecture, utilizing U-net and skip-connections, effectively handles high-resolution features.
• Bi-linear interpolation enabled parameter reduction with no visible degradation in image quality.

Conclusions:

• Deep learning offers an efficient alternative for local dimming, eliminating the need for explicit backlight information.
• The U-net based network with bi-linear interpolation provides a computationally efficient and effective solution for display image compensation.
• This approach advances local dimming technology for displays, enhancing contrast and power efficiency.