Film and Video Quality Optimization Using Attention Mechanism-Embedded Lightweight Neural Network Model

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Video blur caused by camera shake and object motion degrades video quality, obscuring target edges and deforming visuals.
• Existing video deblurring (VD) methods struggle to fully optimize movie video quality.
• Addressing video distortion requires advanced algorithms for clear and accurate visual representation.

Purpose of the Study:

• To propose an advanced video deblurring (VD) algorithm for optimizing movie video quality.
• To enhance feature expression and extraction capabilities in video deblurring.
• To improve the efficiency and effectiveness of video deblurring techniques.

Main Methods:

• A novel video deblurring (VD) algorithm is developed using a neural network (NN) model integrated with an attention mechanism (AM).
• The Haar planar wavelet transform (WT) is employed for video image preprocessing and deblurring in the wavelet domain.
• Spatial and channel attention mechanisms, including residual inception spatial-channel attention (RISCA) and skip spatial-channel attention (SSCA), are fused into the network to extract multiscale features and accelerate training.

Main Results:

• The proposed Haar and attention video deblurring (HAVD) algorithm demonstrated superior performance compared to the multisize network Haar (MSNH) method.
• HAVD achieved improvements of 0.10 dB in peak signal-to-noise ratio (PSNR) and 0.005 in structural similarity (SSIM).
• The integration of dual attention mechanisms significantly enhanced model performance and optimized video quality.

Conclusions:

• The developed Haar and attention video deblurring (HAVD) algorithm effectively addresses video blur and distortion issues.
• The fusion of spatial and channel attention mechanisms is crucial for improving feature expression and overall model performance.
• This research provides valuable technical support for enhancing the quality of movie videos through advanced deblurring techniques.