Modified Local Chan-Vese Model Segmentation And A Distance Function Based Geometrical Approximation Method For Fruit Volume Measurement
Loading...
Date
2015-09
Authors
HANNAN, ZHANG
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Volume is one of the important physical attributes for a fruit since volume is highly related to weight, price and nutrition. Several studies had been conducted on volume measurement by image processing techniques. Image processing is non-destructive and shows good potential in volume measurement. Currently, threshold based methods are commonly used to segment a fruit from the background in an image. Threshold based methods are sensitive to noise, intensity inhomogeneity, shadow, and intensity variation on the fruit surface. For volume measurement, many methods can only give the results with a good accuracy by using multiple images. The geometrical approximation methods can measure the fruit volume by approximating the shape of fruit with a regular and simple geometrical shape. These methods can only maintain their measurement good accuracy for some regular shaped fruits and cannot be applied to fruits with irregular shape. A new energy functional based segmentation algorithm, called modified local Chan-Vese model, is proposed to avoid false segmentation. The proposed segmentation algorithm was tested on the images with heavy noisy, intensity inhomogeneity, shadows and bright spots. Comparing to Otsu threshold method and traditional LCV model, the proposed segmentation algorithm can give a correct segmentation for fruit samples. A new geometrical approximation method is proposed to measure volume for both regular shaped fruits and irregular shaped fruits. In this study, it is assumed that the variation of cross section for fruits followed the property of distance function. The mean absolute percentage error for capsicum, pear and tomato were around 5% and for apple and orange were below 4%. The volume data measured by proposed method were compared with circular approximation and previous studies. The coefficient of determination 2R was 0.925 for apple, 0.976 for capsicum, 0.817 for orange,
0.722 for pear and 0.882 for tomato. The volume measured by proposed method was highly linear correlated to volume measure by water displacement measurement (WDM) for capsicum which has an irregular shape. The results of paired t-test showed that the proposed method was not significantly different from WDM for orange, pear and tomato. The Bland-Altman plots showed that the proposed method had generally better agreement with WDM.
Description
Keywords
Modified Local Chan-Vese Model Segmentation And A Distance Function , Based Geometrical Approximation Method For Fruit Volume Measurement