Defects on product surfaces affect quality of the product. Machine vision provides an efficient tool for the surface defect detection. Threshold is commonly used to separate objects from the image background in the vision-based inspection method. The Otsu method is one of the most used approaches to decide the threshold for a satisfied result when the image histogram is bimodal, but it fails when the histogram of an image is unimodal or close to unimodal. Defects in product surfaces can range from small to large sizes, which results in distributions of the image histogram change from unimodal to bimodal. An improved Otsu method, named the weighted object variance (WOV), is proposed in this research to detect defects on product surfaces. A parameter that equals the cumulative probability of defects occurrence is weighted on the object variance of between-class variance. The weight ensures that the threshold always be a value that locates at the valley of two peaks or at the left bottom rim of a single peak histogram. It is essential to have a high detection rate and low false alarm rate for the defect detection. Experimental results demonstrate the effectiveness of the improved Otsu method in the defect detection of various surfaces. Compared to other thresholding methods such as maximum entropy, Otsu, valley-emphasis, and modified valley-emphasis methods, the WOV method provides better segmentation results. Crown Copyright (C) 2015 Published by Elsevier B.V. All rights reserved.