This paper presents a hybrid framework for detecting and classifying image splicing using a combination of deep learning (DL) and traditional machine learning techniques. The proposed method employs Local Optimal Oriented Pattern (LOOP) to extract spatial texture inconsistencies that help distinguish spliced images from authentic images. These LOOP-derived maps are further refined using a pre-trained ResNet-50 model to capture high-level structural features. A Support Vector Machine (SVM) classifier is then used for the final classification between authentic and spliced images. The method was evaluated on publicly available image forensics datasets, including the Columbia Image Splicing Dataset, CASIA v2.0, and a subset of the Dresden Image Database. All experiments used stratified 5-fold cross-validation, and the reported results represent the mean performance across folds. The proposed framework achieved an average classification accuracy of 95.20%. Instead of claiming general superiority, we compare the method with representative feature-based and deep learning approaches using identical evaluation settings. The results indicate consistent detection performance under common post-processing operations. It shows improved tolerance under moderate blur and scaling, while performance decreases under JPEG compression and geometric transformations. This technique also demonstrates low performance variance across validation folds, indicating stable generalization. By combining local texture descriptors with deep residual features, the framework provides a reproducible and statistically supported approach for image splicing detection. The proposed system contributes to digital image forensics by offering a clearly defined evaluation protocol and measurable performance improvements rather than rhetorical performance claims.