| Abstract [eng] |
The problem of classification of amber is known for a long time among artisans of amber art. Existing automated classification solutions in the past have sorted the amber according to color, but there is a need to expand the possibilities of classifying amber. The proposed solution is able to classify amber by size (optional number of classes), by shape (square, rectangular, oval, circular, trapezoidal, diamond, triangle and asymmetric) and by color. To estimate the size of an object, the number of white pixels that make up the object can be calculated, which can easily be converted to physical unit. This obtained parameter is used as a property for classifying an object by size, which is assigned to predefined classes. Ratings for class assignments are calculated automatically by specifying the desired number of classes and the maximum number of pixels in the object assigned to the highest class. The proposed color classification divides the amber picture into a specified number of segments and evaluates the difference between segments and white color. The algorithm is capable of self-learning. Amber is classified according to the value of tolerance, which reflects the maximum difference between amber segments and white color. A new algorithm was adopted for evaluating the amber form, using additional steps of image processing where the long axis of the object is calculated and rotated parallel to the x coordinate axis. Few more rotation procedures are applied (if necessary), where the narrowest part of the object is aligned to the right with respect to the x axis, and at the top with respect to the y axis. The algorithm used evaluates the length of the object x and y axes passing through the center of the object, the real area of the object, and the actual area of the object, diagonals rotated at a 45-degree angle from those axes and adjacent to the edge of the object. The form of the object is not defined unambiguously, even the expert (human) assignments are not unambiguous. This problem is solved by introducing tolerance values. The resulting classes can be combined - the number of classes is limited to the number of \"pockets\" of the device used for classification. The proposed form classification is up to 11 times faster than using a decision tree ensemble consisting of 3 decision trees. Previous decisions required expert intervention, which had to manually select amber for classes - suggested solution avoids this step. |
