Ensemble Learning for Offline Signature Verification using Fused Deep Features

Signature verification plays an important role in document authentication. The efficient and robust system is required for forgery detection in documents in the presence of distortion such as rotation, scaling and noise. This paper proposes a writer-dependent signature verification method, where hand-crafted features are fused with one-dimensional convolutional neural network features to cater different variance in the whole verification process. Proposed feature extraction methodology is validated by bagged ensemble learning technique in which machine learning classifier’s decision is finalized based on majority voting. The effectiveness of the proposed signature verification approach has been evaluated across three distinct publicly accessible datasets. The outcomes highlight the versatility of this method. Moreover, the proposed system surpasses the performance of other pre-existing systems documented in the literature.