In this paper a new method for optic disc (OD) detection in retinal images has been proposed. It is based on computing the OD location using dictionary based approach. The idea behind the approach is that OD properties within the retinal images remain consistent (i.e. brightness, circular shape and blood vessel convergence at the OD center) and thus OD region can be linearly represented in terms of dictionary elements composed using similar patterns selected from subimage in the training phase. The robustness of OD center detected is ensured by verifying the presence of main blood converges at OD center in its neighbourhood. These main blood vessels are obtained by estimating the background information using large window sized median filter. Also for reducing the average computation time, we have performed OD detection in test image only on bright pixels and assumed pixel intensity as primary component for OD detection. These bright pixels are selected by local thresholding (i.e. 15*15 block) in the test retinal image using Otsu algorithm. The result of proposed method has been obtained on DRIVE, STARE, and DIARETDB1 datasets includes 210 retinal images with success rate of 100%, 95.06% and 98.8%.