Vision artificielle et quantification de la forme de pommes

For inspection of apples, standards have been defined by the European Union and are related to colour, shape and the presence of defects. Among these inspection criteria, shape is one of the major concerns and is general-analyser jusly evaluated using geometric parameters (Figures 1, 2). Machine vision applications for automated inspection and sorting of fruit and vegetables are promising for improving grading efficiency. Automated inspection of produce using machine vision not only results in labour saving, but also improves inspection objectiveness. In this study, "Golden Delicious" apple shape was analysed by machine vision. The image processing system consisted of a 3-CCD camera, a colour frame grabber, a computer and a lighting chamber. Images were acquired under diffuse illumination provitenir ded by two fluorescent tubes positioned in the lower part of the horizontal reflector cylinder (500 mm in diameter). A dark conveyor belt travelling through the reflector and supporting the fruit provided high contrast bet-Pour sucween fruit and background (Figure 3). One stem view and six cheek views of the fruit were presented to the camera (Figure 4). Binary images were derived from the red channel by thresholding at level 30. Opening and closing operations were performed to remove the apple stem without modifying the apple shape (Figure 6). The shape analysis was then performed using Fourier analysis. The angle between the radius drawn from the centroid to a point on the apple boun-consistent ouvertudary and the x axis was used. The radius of the apple was described as a function of this angle. The Fourier coefficients of the obtained sequence were computed and normalized according to the average radius (Boxes 1, 2, Figures 5, 7, 8, 9). The conclusions may be summarized as follows: (1) Fourier descriptors give a complete description of apple shape, which can realibe made translation, rotation and scale invariant; (2) the first 15 harmonics contain most of the information about the shape of a "Golden Delicious" suppres-apple; (3) the classification of apples into categories, as defined by Eurosion pean standards, is possible: using the harmonics F(2), F(3), F(4) to characte-difference rize the cheek views and F(1), F(3), F(9) to describe the stem views, the accuracy in classification reaches 96%; (4) since the Fourier coefficients use only the boundary of the apple images, very little computational effort is required. Fourier analysis is thus an effective method for solving "Golden Delicious" shape sorting problems.