Abstract: Blind pixels of IRFPA consist of fixed bad pixel and random bad pixel, the former is caused by material and manufacture defect, while the latter is mainly caused by temperature drift. Scene-based blind pixel detection and compensation algorithm is the effective method to eliminate these bad pixels and increase image quality. Aiming at the defect that the current filter-based detection methods can't distinguish random blind pixels and weak point targets, the response and noise feature of random blind pixels were first analyzed, and a new adaptive blind pixel detection and compensation algorithm based on fuzzy median and temporal accumulation was proposed. Fuzzy median filter was used to extract the potential blind pixels from scenes, then the exact distribution of fixed and random blind pixels were determined by temporal accumulation, and blind pixel compensation was performed finally. The experiments show that the proposed algorithm can effectively correct bad pixels while avoid misjudging weak point targets.