Worldwide steel industries are rapidly adopting advance data science, AI, ML kind of technologies for increasing interconnectivity and smart automation of their daily processes. As oil and gas companies are seeking higher quality material from steel manufacturer, dependency of above technologies is growing very fast. Currently, defect detection using computer vision is emerging an important technology which is impressing all the technologist and convincing people to accept it. In conventional steel slab caster, various types of internal defects are generated with low to high severity where Centerline Segregation (CLS) and Internal Crack (IC) evolves most significant type of defects, which are likely undesirable. Since, those defects cannot be avoided due to solidification of liquid steel, dynamic soft reduction technology is universally used for minimizing it. Mannesmann standard images is widely used in steel slab caster area for classifying the slab defects severity by comparing the defects printed in sulphur printing paper with the standard images by visual observation to assure material quality. However, this conventional method is highly error prone due to variation in results for varying judgement by operator to operator. Therefore, a suitable scientific method was required to develop for improving reliability of test result. In this study, a quantitative model for classification of CLS and IC defect severity using advance image analytics technique has been developed and described.