Image Segmentation Method for Unbalanced Background in Water Droplets Based on Discrete Cosine Transform and Morphological Processing

Water droplets segmentation is one of the main issues when proposing an automatic method capable to evaluate or classify the surface hydrophobicity in high voltage insulators. In general, two factors pose a challenge in the process of segmenting water droplets images: irregular or unbalanced brightness on the background, due to lighting conditions, and the transparency of water which cause the color levels of the object and background to be fairly close to each other. This paper presents an effective method to segment water droplets images with unbalanced background based on the discrete cosine transform. The proposed method uses the discrete cosine transform to reconstruct the background image regardless of any brightness disturbances and then uses image differentiation and morphological processing techniques to segment the water droplets from the background. When compared to other techniques, the proposed method diminishes the impact of irregular lighting while reducing the computational effort demanded by methods that rely either on processing the image at the HSV color space or on edge detection and processing algorithms. Experimental results on an image database containing water droplets of multiple sizes and shapes demonstrate that the proposed method is capable of successfully segment the expected objects. Thus, the algorithm described in this paper archive similar or better results, for this particular dataset, than the other analyzed techniques.