Lower back pain (LBP) is widely prevalent in people all over the world and negatively affects the quality of life due to chronic pain and change in posture. Automatic localization of intervertebral discs from lumbar MRI is the first step towards computer-aided diagnosis of lower back ailments. Till date, most of the research has been useful in determining a point within each lumbar disc, hence we go one step further and propose a localization method which outputs a tight bounding box for each disc. We use HOG (Histogram of Oriented Gradients) features along with SVM (Support Vector Machine) as classifier and successfully combine these machine learning techniques with heuristics to achieve 99% disc localization accuracy on 53 clinical cases (318 lumbar discs). We also devise our own metrics to evaluate the accuracy and tightness of our disc bounding box and compare our results with previous research.