Techniques of Analysing the Colour of Produces for Automatic Classification Marius Buzera 1 , Voicu Groza 2 , Gabriela Prostean 3 , Octavian Prostean 4 1 Colegiul National Tehnic "Gheorghe Magheru" Str. Lt. Col. Dumitru Petrescu nr.3, Targu-Jiu, Romania 2 University of Ottawa 800 King Edward Avenue, Ottawa, K1N 6N5, Canada 3,4 “Politehnica” University of Timisoara Blvd.V.Parvan, no.2, 300223, Timisoara, Romania Abstract — In the classification of fruits and vegetables, colour is one of the most important parameter that allows for the evaluation of their degree of maturity and freshness, and existence of faults. Also, colour along with its level of homogeneity influences the degree of acceptance of consumers, as well as the pricing. Qualitative sorting is usually performed by trained inspectors. This type of evaluation is rather expensive and is determined by operators’ inconsistency and subjectivity. Machine vision technology offers objective solutions for all these problems and it is considered to be a promise for replacing the traditional human inspection methods. This paper introduces and discusses several techniques of analysing the colour of produces, based on image processing. An automatic video- inspection system was conceived, designed and realised, and it was employed for testing these techniques. The details regarding its design and the conclusions resulted from these experiments are presented here, as well. I. INTRODUCTION „Machine vision” technology firstly came into use in the sixties with such operations as robot manipulations or object classification, while its success was being proved in various fields, namely: inspecting and packing products, inspecting the labels of products, analyzing terrestrial images and examining the moon surface, identifying and tracking military targets, medical diagnosis, autonomous vehicles and robots handling. This technology has lately reached its highest in particular domains, like analysis and classification of some biologically originated produces. The fact that most of these classification systems are used in the industrial area, where the sorted products have a very rigid structure, brings about errors if they are used for sorting produces. Vegetable produces are unique in nature and they develop in accord with environmental conditions that may cause irregularities and variations. Thus, for such categories, the sorting systems have to evaluate the produces’ parameters according to features with a very large range of variation. [1] Studies have pointed out that mechanical techniques, which researchers used for object classification, often caused higher degree of damage, while the accuracy was relatively low. Also, these techniques could not be extended to colour classifications of produces, nor to the detection of their faults. In 1988, Mayers presented in his paper [12], both the advantages and disadvantages of employing human operators in the process of produces’ classification and inspection. He pointed out the advantages of placing human operators before the working font, instead of placing them on the sides, as well as the disadvantages of using hand labour. In his paper, in 1994, Deck [7] spoke about both the advantages and disadvantages of a semi automatic sorting system. Consequently, considering the features of vegetable produces, one can state that the „machine vision” technology is the only technology that can be possibly used. It enables scoring good results, and, at the same time, its lowering prices makes it more and more attractive for general use. Plus, some of these techniques, i.e., non-invasive evaluation methods, allow for the analysis of brittle products as well. Another advantage they offer is the great number of parameters that they can analyse, such as: shape, colour, size, homogeneity of colours, existence of faults, etc. Although these parameters are not difficult to be comprehended and assessed by humans, they are rather difficult to be evaluated and quantified by computers. This paper presents several analysis algorithms for classification of produces by colour, along with a framework that was developed to test them; this inspection and classification system runs these algorithms in real time, while, following the principles of „machine vision” theory. The developed system was employed for testing the shape and the colour of some industrial products and vegetable produces, and the achieved results were promising. Since both the colour of industrial products and that of vegetable produces are prone to changes, one should observe an intermediary stage of interactively setting the range of colour variations according to each class, before starting a specific sorting process. The developed application, presented in this paper, allows for interactive setting of the ranges of colour variations which are specific for each class, as well as changing the position of both the video inspection system, and the light intensity of the illumination system. 978-1-4244-2083-4/08/$25.00 ©2008 IEEE