Harumanis Mango Leaves Image Segmentation based on Wavelet Transformation with Phansalkar and Sauvola Thresholding

Abstract

Harumanis mango is one of the economic sources of the Perlis state. It has a sweeter taste compared to other mangoes. However, the Harumanis mango tree required specific weather, soil nutrient contents and pH level. This makes the farmer does not know the health condition of their Harumanis mango tree. Therefore, this project aims to provide the best method of leaves detection to the farmer. The leaves image samples are collecting from the orchard and undergo pre-processing. Then the input image was converted into grayscale with principal component analysis (PCA). Wavelet transformation was implemented to increase the discriminability of the segmentation technique for separating the leaf and background. The leaf segmentation is done by using Phansalkar and Sauvola thresholding techniques. After that, fill hole and area opening techniques are implementing to reduce noise in the image. These two thresholding techniques are comparing and discuss with their segmentation performance. Overall, Phansalkar thresholding has produced better performance in segmenting healthy and unhealthy Harumanis mango leaves with sensitivity, specificity and accuracy of 92.05%, 81.37% and 83.51%, respectively.