In the field of fingerprint identification, local histograms coding is one of the most popular techniques used for fingerprint representation, due to its simplicity. This technique is based on the concatenation of the local histograms resulting in a high dimension histogram, which causes two problems. First, long computing time and big memory capacities are required with databases growing. Second, the recognition rate may be degraded due to the curse of dimensionality phenomenon. In order to resolve these problems, we propose to reduce the dimensionality of histograms by choosing only the pertinent bins from them using a feature selection approach based on the mutual information computation. For fingerprint features extraction we use four descriptors: Local Binary Patterns (LBP), Histogram of Gradients (HoG), Local Phase Quantization (LPQ) and Binarized Statistical Image Features (BSIF). As mutual information based selection methods, we use four strategies: Maximization of Mutual Information (MIFS), minimum Redundancy and Maximal Relevance (mRMR), Conditional Info max Feature Extraction (CIFE) and Joint Mutual Information (JMI). We compare results in terms of recognition rates and number of selected features for the investigated descriptors and selection strategies. Our results are conducted on the four FVC 2002 datasets which present different image qualities. We show that the combination of mRMR or CIFE feature selection methods with HoG features gives the best results. We also show that the selection of useful fingerprint features can surely improve the recognition rate and reduce the complexity of the system in terms of computation cost. The feature selection algorithms may reach 98% of time reduction by considering only 20% of the total number of features while also improving the recognition rate of about 2% by avoiding the curse of dimensionality phenomena.