The term ’saliency’ refers to a visual characteristic of an image where the significant regions are immediately recognizable. While deep learning approaches excel in various domains, our proposed methodology is designed for scenarios where the drawbacks of deep learning pose substantial hindrances. Our methodology addresses the limitations of deep learning, such as extensive training and data scarcity, by using feature extraction methods that avoid prolonged training times and focus on ensembling. The proposed approach is a three-step method that integrates different contextual features and objectness. In the first step, the input image is segmented into over-segmented regions using a simple linear iterative clustering algorithm. From this segmentation, a local saliency map is constructed by considering both color and spatial cues. Moving to the second step, an objectness map is generated using object proposals, aiding in identifying superpixels belonging to the foreground object. The saliency map generated in this step distinguishes salient objects from the background. The third step involves obtaining a thresholded saliency map by applying a watershed algorithm to the input image, followed by calculating a minimum directional backgroundness score. The saliency maps generated in these steps are then combined to yield the final saliency map. To evaluate the proposed algorithm, extensive testing is performed on two publicly available datasets. The algorithm is compared with ten bottom-up approach methods. The proposed method demonstrates its strength in efficiently detecting salient objects in various image categories, including Center-focused, low contrast, complex background, boundary-focused, and multiple object scenarios. The algorithm’s performance is assessed using standard metrics commonly used in salient object detection evaluations, including precision (0.88), recall (0.91), F-measure (0.86), and mean absolute error (MAE) of 0.12 on the ECSSD dataset.