Usually, in a real-world scenario, few signature samples are available to\ntrain an automatic signature verification system (ASVS). However, such systems\ndo indeed need a lot of signatures to achieve an acceptable performance.\nNeuromotor signature duplication methods and feature space augmentation methods\nmay be used to meet the need for an increase in the number of samples. Such\ntechniques manually or empirically define a set of parameters to introduce a\ndegree of writer variability. Therefore, in the present study, a method to\nautomatically model the most common writer variability traits is proposed. The\nmethod is used to generate offline signatures in the image and the feature\nspace and train an ASVS. We also introduce an alternative approach to evaluate\nthe quality of samples considering their feature vectors. We evaluated the\nperformance of an ASVS with the generated samples using three well-known\noffline signature datasets: GPDS, MCYT-75, and CEDAR. In GPDS-300, when the SVM\nclassifier was trained using one genuine signature per writer and the\nduplicates generated in the image space, the Equal Error Rate (EER) decreased\nfrom 5.71% to 1.08%. Under the same conditions, the EER decreased to 1.04%\nusing the feature space augmentation technique. We also verified that the model\nthat generates duplicates in the image space reproduces the most common writer\nvariability traits in the three different datasets.\n