Cardiovascular diseases (CVDs) are a leading cause of death every year. Timely detection is critical for reducing CVD-related fatalities. We use an artificial intelligence (AI) method that uses spectrogram images of heart sounds. Spectrograms are a visual representation of audio signals and are compact and efficient. We propose a novel supervised machine learning (ML) model that is lightweight and uses the Histogram of Oriented Gradients (HOG) and Local Binary Patterns (LBP) textural features of the spectrograms. These two features and a combination of these are then given as input to ML classifiers. Our methods performed well across a dataset consisting of 957 heart sound recordings including one normal and four different types of heart diseases. The binary classification task distinguishing normal from abnormal cases attains a high accuracy of 99.78%. For the more challenging five-class classification, our approach achieves 98.62% accuracy. The high values of F1-score, precision, and recall values indicate that the method is reliable. The fusion of HOG and LBP features with the kNN classifier has given exceptional results. Our model gives excellent results even when faced with imbalanced data and was executed in a low-end system in very little time. Our AI-driven diagnostic method exhibits immense potential for the early detection of heart diseases. Its accuracy, efficiency, and noninvasive nature make it useful for diverse healthcare settings.