Digitalization of handwritten content is an essential process to process and store for a long time. Recognition of characters’ form printed and handwritten material used for digitalizing handwritten content. Digitalization of handwritten characters used to convert into different languages and helps to store over a longer time advanced deep learning technologies are more effective learning models for character recognition. To recognize Gujarati characters from handwritten materials, proposed study contributes an ensemble learning model for blur and joint character recognition. InceptionV2 and ResNet50-based networks are used to optimize feature extraction. Proposed study uses MLP layer for the final classification of Gujarati handwritten characters. Simulation of proposed study uses two datasets as separate digit and Gujarati characters. Proposed ensemble learning model is able to achieve 98.47 and 99.01% accuracy for character and digit recognition, respectively. Simulation of the study also compares performance of different deep learning models and impact of final classification layers. Simulation of the study also tested with various parameters for training the model for Gujarati character recognition.