Textile industry is one of the noticeable contributors to our nation’s growth. The quality control procedures in textile production primarily involves the defect detection process. For detecting the defects in complex fabric textures, proper construction of sparse representation is needed. Existing fabric defect detection methods are incapable of detecting defects in more than one type of fabric and have increased detection time while missing few defects. In this paper, dictionary learning is proposed which is used to learn the sparse representation of complex data. Three types of greedy algorithms OMP, ROMP and STOMP are used for sparse representation and the results are compared based on computational speed and accuracy. The experimental results indicate that the STOMP algorithm gives accurate and precise results with lesser time consumption. STOMP achieves 99.3% reduction in time consumption compared to OMP and 97.7% reduction in time consumption compared to ROMP. Also, if ROMP and STOMP are used for signal recovery, the formulation of joint matrix is not essential resulting in reduced computational complexity.