The emergence of increased sources for Big Data through consumer recording devices gives rise to a new basis for the management and governance of public infrastructures and policy de-sign. Road maintenance and detection of road surface defects, such as cracks, have traditionally been a time consuming and manual process. Lately, increased automation using easily acquirable front-view digital natural scene images is seen to be an alternative for taking timely maintenance decisions; reducing accidents and operating cost and increasing public safety. In this paper, we propose a machine learning based approach to handle the challenge of crack and related defect detection on road surfaces using front-view images captured from driver’s viewpoint under diverse conditions. We use a superpixel based method to first process the road images into smaller coherent image regions. These superpixels are then classified into crack and non-crack regions. Various texture-based features are combined for the classification mod-el. Classifiers such as Gradient Boosting, Artificial Neural Network, Random Forest and Linear Support Vector Machines are evaluated for the task. Evaluations on real datasets show that the approach successfully handles different road surface conditions and crack-types, while locating the defective regions in the scene images.