Deep learning-based writer verification methods have achieved high accuracy in recent years, primarily by relying on segmented inputs such as individual words or lines. However, this segmentation-based approach can be problematic in real-world contexts such as examination scenarios where handwriting often displays significant intra-writer variability due to stress, time pressure, and writing speed. In such cases, precise segmentation becomes difficult and may degrade verification performance. To address these challenges, this work presents a texture-based deep learning approach for offline, text-independent writer verification. Instead of relying on segmented characters or word-level inputs, the handwritten documents are converted into texture images that emphasize global writing style. These representations are processed using a Siamese Neural Network with a Vision Transformer (ViT) feature extractor. The model learns discriminative embeddings between paired texture patches through contrastive loss, enabling it to distinguish between different writers effectively. Evaluation is conducted on a privately collected examination-style dataset containing 600 samples from 100 writers, as well as the CVL and IAM datasets. Across all test sets, ViT achieves consistent performance, with ROC-AUC scores exceeding 94.8% and accuracy around 87.9%. Results indicate texture-based representations constitute a viable and effective substitute for conventional verification processes that rely solely on line or word segmentation. This approach generalizes well by focusing on writing style over content, making it suitable for real-world writer verification.