Background: The knowledge of subcellular location of proteins is essential to the comprehension of numerous protein functions.

Objective: Accurate as well as computationally efficient and reliable automated analysis of protein localization imagery greatly depend on the calculation of features from these images.

Methods: In the current work, a novel method termed as MD-LBP is proposed for feature extraction from fluorescence microscopy protein images. For a given neighborhood, the value of central pixel is computed as the difference of global and local means of the input image that is further used as threshold to generate a binary pattern for that neighborhood.

Results: The performance of our method is assessed for 2D HeLa dataset using 5-fold crossvalidation protocol. The performance of MD-LBP method with RBF-SVM as base classifier, is superior to that of standard LBP algorithm, Threshold Adjacency Statistics, and Haralick texture features.

Conclusion: Development of specialized systems for different kinds of medical imagery will certainly pave the path for effective drug discovery in pharmaceutical industry. Furthermore, biological and bioinformatics based procedures can be simplified to facilitate pharmaceutical industry for drug designing.

背景：蛋白质亚细胞定位的知识对于理解多种蛋白质功能至关重要。

目的：蛋白质定位图像的准确以及计算有效，可靠的自动化分析在很大程度上取决于这些图像的特征计算。

方法：在当前的工作中，提出了一种称为MD-LBP的新方法，用于从荧光显微镜蛋白质图像中提取特征。对于给定的邻域，将中心像素的值计算为输入图像的全局均值和局部均值之差，并进一步用作生成该邻域的二进制模式的阈值。

结果：使用5倍交叉验证协议对2D HeLa数据集评估了我们方法的性能。以RBF-SVM为基础分类器的MD-LBP方法的性能优于标准LBP算法，阈值邻接统计和Haralick纹理特征。

结论：针对不同类型医学影像的专业系统的开发必将为制药行业有效发现药物铺平道路。此外，可以简化基于生物和生物信息学的程序，以促进制药业进行药物设计。