Object segmentation for fruit images using OHTA colour space and cascade threshold

Object segmentation is a key step in image analysis and the most difficult low-level image analysis tasks, specifically in the semantic objects approach. This approach is widely implemented in various domains including fruit images. Object segmentation using OHTA colour space is one of successful methods to separate fruit object and background. However, the method is prone to remove shadows or other noises. Therefore, this study proposed an extended method of OHTA-based object segmentation to overcome those problems by applying cascade threshold. The selected threshold value is 50. It is based on the optimal threshold value to reduce both of over-segmented and under-segmented images. The proposed method increases the overall accuracy of about 41.25%, 52.5%, and 20% for tomato images, apple images, and banana images respectively.