In this project, the numerical color vision is tested parallel to IRM and CT as new reference method to grad pig carcasses and to predict lean meat percentage in relation with EU reference dissection. A vision system and image analysis are used by Cemagref to study various cuts on the carcass for the prediction of lean meat percentage.
These cuts have been defined by the experts of the pig carcasses grading of Eupigclass project. 10 cuts are made by a butcher on the 3 joints (4 in the loin, 2 in the ham, 3 in the belly and 1 between the loin and the belly) according the EU reference dissection method on the half carcass. During the first trial 60 half carcasses are cutted and studied. The split line is also studied.
A vision system has been devised in France by the Cemagref and was installed and ajusted in the Diagnostic Center in Kaposvar University. This system has got two 3CCD color cameras, a board of acquisition and numerisation of images on a PC, a uniform and indirect lighting system, a post to hang the half carcass and a inclinable and removable table for the cuts of pieces of meat, a set of patches and a grey uniform background for color calibration.
15 color images of 10 cuts on the carcass are grabbed with the vision system in RGB (Red, Green, Blue) color space of the CCD Cameras. Images of a grey background and color standard patches are also grabbed for color calibration.
To compare image and to make the same image processing on each image a color calibration is realized to correct instability in time and defect of geometrical uniformity of the vision system. A transformation of color space is executed to obtain L*a*b* images beacause L*a*b* components are decorrelated and can be studied independently.
To separate the various tissues, the histograms of tissues images are studied and image processing algorithms are devised to segment each tissues (Fat, muscle, bones). Currently Cemagref is developping algorithms to segment various tissus in the L*a*b* color space in these images .
On segmented images of tissues, measurements like areas, thickness will be made on fat and muscle. These measurements are compared on each cuts , between cuts and will be used for the predicting of the lean meat percentage. The choice of measurements is important for the model even though we have to anticipate and to make choice of these measurements a priori without knowledge.