The document describes a technique for detecting and removing defocus blur from images using a convolutional neural network (CNN) and local metric map. The CNN is first trained on a publicly available blur image dataset to detect blurry regions. Then, a local metric map analyzes each pixel's intensity to identify blurry versus sharp areas. Pixels with high intensity are kept, while low intensity pixels in the blurry regions are removed. This process converts the image to a clean version without blur. Screenshots show the software interface for training the CNN model, uploading test images, and viewing the results of blur detection and removal.