Among several diseases, COVID-19 has highlighted the urgent need for AI-based tools o develop precise, accurate, and effective containment and treatment plans. This paper utilizes chest images and introduces a hybrid architecture tailored for COVID-19 detection. The proposed architecture integrates multiple modules to extract texture and shape cues, thereby facilitating more accurate and discriminative classification. The analysis pipeline starts with an auto-encoder-based input scaling module to filter out noise and to keep prominent image features. Subsequently, two sets of features are extracted from the preprocessed images representing rich concentration of local and global features. We integrated various texture descriptors (histogram of oriented gradients, gray-level co-occurrence matrix, and local binary patterns) with deeper features extracted from an optimized neural architecture. Thirdly, a feature fusion and weighing module is used to weigh the feature sets to produce an optimal feature vector for classification using a cascaded machine classification module. Evaluation of the proposed methods on publicly available datasets using various metrics and ablation studies documents the robustness of our weighted ensemble classification system. Cross-validation evaluation demonstrate higher accuracy of the proposed system with an average accuracy of 96.5% and 95.5% on two datasets and when compared against other recent methods.