Blur image detection and classification using edge detection techniques

Abstract

Blur classification is important for blind image restoration. It is a challenging task to detect blur in a single image without any information. In this paper, we used edge detection techniques and deep learning convolutional neural network named ResNet 50 for the classification of blur-type images. ResNet 50 model effectively reduces gradient disappearance problem and uses skip connection to train the dataset. Generally, images are subjected to defocus and motion blur, which is caused by the improper depth of focus and movement of objects at the time of capture. Kaggle's blur dataset is used in this paper, which consists of sharp, defocus and motion blur images. Edge detection techniques are applied on images using Laplacian, Sobel, Prewitt, and Roberts filters to obtain features like mean, variance, maximum signal-to-noise ratio (SNR), which are used to train the system and classify the images using classification algorithm.