Diabetic Retinopathy (DR) classification is approached with variety of techniques. The methods rely on the set of features like color, shape, edge, handicraft, and binary patterns. Still, the methods suffer to achieve higher performance as they lack to consider layered features. With the motivation to improve DR classification accuracy, an layer centric color approximation segmentation and capsule network-based diabetic retinopathy classification (LCAS-CN) model is presented in this article. The fundus images are initially preprocessed with Level Based Normalization algorithm which removes noise and enhance the quality. Further, Multi-Level Color Approximation (MLCA) algorithm is applied towards segmenting the image. The capsule network used to train the images with the features of data set. The method removes black spots, ruptures, edges, hemorrhages, wool spots of cotton, exudates, abnormality in vessel growths, etc. During the training phase, it identifies and takes out the image’s texture and other features to generate the capsule. The capsules at the middle layer mainly used to compute the weight calculation to give assistance for data classification. The capsules are intended to calculate different disease influence values towards various factors. According to the disease influence value of multiple factors, the high-level capsule computes the Diabetic Retinopathy Support (DRS) against various classes. According to the value of DRS, the method identifies the class of the image given. The proposed MMCAS-CN introduces higher accuracy in disease classification with less time complexity.