Optical character recognition (OCR) is a widely used technology to convert text images to editable text. Researchers already proposed many machine learning algorithms to address this problem. However, Bangla text recognition is highly challenging job for its complicated writing style, compound characters and highly diversified fonts. To address the segmentation problem we have proposed an algorithm namely Blob-Labeled character Segmentation (BLCS) that initiates with an extensive preprocessing to extract the characters from text. Our novel character segmentation procedure extracts characters maintaining 97.5% accuracy. Unsupervised feature learning becomes a powerful tool in machine learning nowadays. To increase the recognition rate of the characters, we have introduced a fuzzy unsupervised feature learning algorithm to learn features of individual characters. We then use Artificial Neural Network (ANN) and Support Vector Machine (SVM) to classify the characters. The SVM provides 99.4% accuracy which outperforms all other approaches.