A system and method for detection of a fraudulent action using face database search and retrieval is given. The method identifies fraudulent action using a combination of images captured in real-time, feature vector extraction, and metadata matching. The method involves capturing, a query selfie image and a query ID card image. Thereafter, face feature vectors from the query images are extracted, and candidate face feature vectors are identified by searching a set of nearest neighbors of the extracted face feature vectors. The method then retrieves pre-stored application IDs associated with the candidate face feature vectors and a pre-stored metadata associated with the candidate face feature vectors. The method then matches the query metadata with the pre-stored metadata and generates either a fraudulent or non-fraudulent indication based on the matching. Thereafter it detects a fraudulent activity based on the fraudulent indication.