Scholarlink

<p>Writer identification from offline handwriting images is an active research problem that has garnered significant attention in recent years. Several approaches based on the study of text structure, texture, or deep learning have been proposed to map each manuscript to its likely writer. In this article, we propose a new texture descriptor, the Binary Neighbors Model (BNM), as the basis for a writer recognition system. Inspired by the Local Binary Pattern (LBP), the BNM descriptor compares neighboring points of each image point against a fixed threshold that varies depending on the context of the recognition task. The proposed system commences with the extraction of local features from BNM histograms of small patches extracted around FAST key points. These features are then encoded using the Vector of Locally Aggregated Descriptors (VLAD) method and classified employing the K-Nearest Neighbors (KNN) technique. The efficacy of the newly developed BNM descriptor is validated through comparative analysis with two established texture descriptors, Local Binary Pattern (LBP) and Local Phase Quantization (LPQ). Additionally, an experimental study was conducted on seven public databases: IAM, CVL, KHATT, Firemaker, CERUG-CN, CERUG-EN and ICDAR 2017. The study yielded successful identification rates of 97.9%, 100%, 97.1%, 98.4%, 100%, 100% and 77.4%, respectively. These results demonstrate the promising potential of the BNM descriptor for writer recognition.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Binary Neighbors Model (BNM) : An LBP-Inspired Descriptor for Offline Writer Identification

  • 期刊论文

  • 发表日期：

    2025年6月13日

查看全文
  • Abdelillah Semma,
  • Said Lazrak,
  • Nabil Lamii,
  • Youssef Malhouni,
  • Yaâcoub Hannad

摘要

Writer identification from offline handwriting images is an active research problem that has garnered significant attention in recent years. Several approaches based on the study of text structure, texture, or deep learning have been proposed to map each manuscript to its likely writer. In this article, we propose a new texture descriptor, the Binary Neighbors Model (BNM), as the basis for a writer recognition system. Inspired by the Local Binary Pattern (LBP), the BNM descriptor compares neighboring points of each image point against a fixed threshold that varies depending on the context of the recognition task. The proposed system commences with the extraction of local features from BNM histograms of small patches extracted around FAST key points. These features are then encoded using the Vector of Locally Aggregated Descriptors (VLAD) method and classified employing the K-Nearest Neighbors (KNN) technique. The efficacy of the newly developed BNM descriptor is validated through comparative analysis with two established texture descriptors, Local Binary Pattern (LBP) and Local Phase Quantization (LPQ). Additionally, an experimental study was conducted on seven public databases: IAM, CVL, KHATT, Firemaker, CERUG-CN, CERUG-EN and ICDAR 2017. The study yielded successful identification rates of 97.9%, 100%, 97.1%, 98.4%, 100%, 100% and 77.4%, respectively. These results demonstrate the promising potential of the BNM descriptor for writer recognition.