With the widespread application of biometric, unprotected biometric templates is still at risk of serious security and privacy breaches. Existing biometric template protection schemes destroy the structure of the original biometric to satisfy irreversibility, reduce the recognition accuracy of biometric templates, and often face challenges in balancing accuracy and security. Therefore, a cancelable fingerprint template protection scheme based on random projection with improved Min-Max hash is proposed to alleviate the balance problem of accuracy and security while achieving fingerprint template revocability. First, the Gaussian random projection matrix is used to randomly project the high-dimensional original fingerprint features into low-dimensional space to maintain the distance similarity between the low-dimensional features and the original fingerprint features. Then, the Min-Max hash algorithm is improved by using the minimum index, subminimum index, sub-maximum index, and maximum index of the projected low-dimensional features as the hash code. This improvement alleviates the reduced recognition accuracy caused by the deviation of the minimum and maximum indexes in the Min-Max hashing algorithm while enhancing the efficiency of generating fingerprint templates. Finally, a cross-matching method based on Jaccard similarity is proposed to increase the collision probability between hash codes, further improving the recognition accuracy of fingerprint templates. Experimental results and analyses show that this scheme improves the recognition accuracy and generation efficiency of fingerprint templates while resisting security and privacy attacks. This approach demonstrates high universality for the protection of biometric templates such as fingerprints and faces.