A CNN-based Blurriness Measure Method for Focused Object Segmentation

Aiming at blurriness measure, we propose a convolutional neural network (CNN)-based method integrated with regression. Examining 16 VGG-like models, we select the proper model to compute pixel blurriness. The regression target is the computational measure by referring to the previous work. Subsequently, the thresholding applies to the pixels’ blurriness, and then the area consisting of the selected pixels with small blurriness is considered to the focused object. The experiment results demonstrate two accomplishments: our method generates similar blurriness to ground truth and achieves higher performance than UNet in focused object segmentation.