Early detection of Autism Spectrum Disorder (ASD) is crucial for effective intervention. This study proposes a machine learning framework for ASD detection using non-invasive facial image analysis. We leverage oriented Basic Image Features (oBIFs) as texture descriptors to capture local structural patterns in facial images. These features are processed through classical machine learning classifiers, with Support Vector Machines (SVM) demonstrating superior classification performance. Our approach achieves high diagnostic accuracy while maintaining model interpretability—a critical factor for clinical adoption. The method offers a scalable, cost-effective alternative to traditional diagnostic protocols. Future research will investigate multimodal integration (speech, behavioral cues, and physiological signals) and real-time screening applications.