CSPP-IQA: a multi-scale spatial pyramid pooling-based approach for blind image quality assessment.

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Traditional Convolutional Neural Network (CNN) based Image Quality Assessment (IQA) methods require images to be resized, often distorting content and reducing accuracy.
• Limitations in feature size of fully connected layers in CNNs necessitate image pre-processing, impacting the integrity of the original image data.

Purpose of the Study:

• To propose a novel blind image quality assessment method, CSPP-IQA, that avoids image resizing.
• To enhance the accuracy, generalization, and efficiency of image quality assessment compared to existing methods.

Main Methods:

• Developed a blind image quality assessment method (CSPP-IQA) utilizing multi-scale spatial pyramid pooling.
• Integrated a convolutional block attention module and an image understanding module within the CSPP-IQA framework.
• Enabled direct input of original images without pre-defined size adjustments.

Main Results:

• CSPP-IQA demonstrated superior accuracy, generalization, and efficiency over traditional IQA methods.
• Experiments on real-scene IQA datasets validated the effectiveness and efficiency of the proposed CSPP-IQA method.
• The multi-scale spatial pyramid pooling approach effectively handles variable input image sizes.

Conclusions:

• CSPP-IQA offers a significant advancement in blind image quality assessment by preserving original image structure.
• The method provides a more robust and accurate solution for evaluating image quality without pre-processing constraints.
• The integration of attention and understanding modules contributes to improved IQA performance.