Edge Detection plays a vital role in machine vision applications and thereby variety of edge detection algorithms being developed over time for both grey scale and colour images. In this paper, a new technique for edge detection called cumulative mean intensity differential transition algorithm (CuMIDT Algorithm) is proposed. This approach focuses on learning variations in the local pixel intensities and predicting the possible edge when the intensity deviation goes out of the stipulated window area. Ramps at the edge boundaries and zero crossing are addressed using differential transition model. Experimentation are done on standard FDDB dataset and real dataset. It is observed that the proposed approach gives better results when compared to the recently proposed novel edge detection algorithms.