Scholarlink

In the face of escalating cybersecurity threats targeting Android devices, developing accurate and explainable malware detection methodologies has become increasingly crucial. This paper studies the relevance of using image representation of bytecode for Android malware detection by introducing a new metric, the Median Color Proximity (MCP). The MCP quantifies the degree to which applications with similar behaviors are represented by visually similar images, which we argue is a key prerequisite for using image-based representation of Android Application Packages (APK). We conducted our analysis on a dataset of 29,555 APK files, employing Uniform Maniform Approximation and Projection (UMAP) for clustering on both their behavior and their visual representation. Our findings reveal a significant correlation between pixel distribution and application behavior, suggesting that image-based representations effectively capture the essential behavioral characteristics of APKs.

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

Relevance of Imaged-Based Representation for Android Malware Detection

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Foucauld Estignard,
  • Adrien Djebar,
  • Hugo Deduit,
  • Sourav Rai,
  • Adam Talbi,
  • Christophe Rodrigues,
  • Nga Nguyen

摘要

In the face of escalating cybersecurity threats targeting Android devices, developing accurate and explainable malware detection methodologies has become increasingly crucial. This paper studies the relevance of using image representation of bytecode for Android malware detection by introducing a new metric, the Median Color Proximity (MCP). The MCP quantifies the degree to which applications with similar behaviors are represented by visually similar images, which we argue is a key prerequisite for using image-based representation of Android Application Packages (APK). We conducted our analysis on a dataset of 29,555 APK files, employing Uniform Maniform Approximation and Projection (UMAP) for clustering on both their behavior and their visual representation. Our findings reveal a significant correlation between pixel distribution and application behavior, suggesting that image-based representations effectively capture the essential behavioral characteristics of APKs.