Image classification is a fundamental technique in Computer vision that is used to detect and classify objects and is widely used in applications such as autonomous vehicles, surveillance systems, biological species classification, online shopping image categorization, etc. Although convolutional neural network (CNN) models have proven to be the best supervised image classifiers with a high rate of accuracy, they often struggle to capture essential features and structural information from the images and effectively utilize the global statistical information, leading to suboptimal classification accuracy, especially when the number of classes increases. This study identifies such limitations and focuses on improving the performance of image classification within CNN architectures by integrating histogram features. We use the Histogram of Oriented Gradients (HOG) technique to extract histogram feature descriptors and an oriented gradient image from each train image, as well as from the enhanced train image datasets, to integrate them as an additional set of inputs into the CNN model along with the train image tensor data. By incorporating histogram features, CNN architectures aim to improve their ability to capture and leverage global statistical information to enhance their discriminative power for more accurate, reliable, and improved image classification across various datasets and image classes in real-world applications.