Automation in Blur Image Detection with Segmentation using Machine learning

Abstract: One of the definition of blurring with respect to Laplacian is decrease in the variance. Blurring with respect to frequency domain means low frequency where high frequency represents edge. It occurs either because of motion or out of focus parameter. Blur is classified in two types (Local and Global Blur). In this research paper, both the blurred and unblurred images are passed through Fourier transform which calculates the high frequencies as well as low frequencies from the images. In other words, it transforms image from spatial domain to frequency domain. Labels for both blurred as well as unblurred images are chosen as the target i.e. 1 for blurred and 0 for unblurred. These set of images are trained using both support vector machine as well as logistic regression which are tested on the real time images which detects the blur from the images. Trained Model run over the real time image and captures the blur from it. Both state of art method i.e. support vector machine as well as logistic regression is compared in the terms of performance parameters and it is concluded from the results that later winds over the former in both accuracy as well as receiver operating characteristic curve.

Keywords: Support vector machine, Logistic regression, binary values, receiver operating characteristic curve