Research on Non-reference Text Image Blur Assessment System

The non-reference image blur assessment (NR-TIBA) system is significant in text image processing. Due to the high cost of subjective evaluation and the unavailability of reference images in the natural environment, an objective NR-TIBA system is essential. Quantitatively, some traditional methods are affected by the richness of image content, while deep learning methods are too expensive to label; qualitatively, research on text image blur evaluation is scarce. This paper proposes a simple evaluation system for quantitative analysis using deep learning qualitative analysis and the combination of the spatial domain and frequency domain. It has good quantitative performance and only requires a small amount of complex labeling data. Experiments show that the proposed system performs well and can be further applied.