<p>This paper addresses the offline handwritten signature recognition task, a challenging classification problem due to high intra-class variability and inter-class similarity. The objective is to accurately classify signature images into their corresponding identity classes, despite inherent uncertainty and imprecision in signature data. To address this, a convolutional neural network is employed for feature extraction, generating a robust representation of each signature-a discriminative and noise-tolerant feature vector that captures essential structural elements of the signature. These features are then classified using a Fuzzy Min-Max Neural Network, which excels in handling overlapping and ambiguous patterns. However, Fuzzy Min-Max Neural Network’s performance is highly sensitive to its parameters (vigilance threshold and sensitivity), often requiring time-consuming manual tuning. To overcome this limitation, an Evolution Strategy is proposed for automatic parameter optimization. This method uses an adaptive fitness function and a dynamic mutation strategy to efficiently explore the parameter space and accelerate convergence. The approach is validated on a proprietary offline signature dataset, and the results demonstrate that the proposed Evolution Strategy-Fuzzy Min-Max Neural Network system achieves superior accuracy, precision, and robustness compared to baseline and state-of-the-art methods.</p>