Intramuscular Fat Percentage Estimation through Ultrasound Images Jos´ e Luis Nunes, Alicia Fernandez, and Federico Lecumberry Facultad de Ingenier´ ıa, Universidad de la Rep´ ublica Julio Herrera y Reisig 565, 11300, Montevideo, Uruguay jlnunes@gmail.com, {alicia,fefo}@fing.com http://iie.fing.edu.uy Abstract. This work presents a framework to estimate intramuscular fat percentage on live cattle based on ultrasound images. A procedure to automatically determine the region of interest is proposed. Given the determined ROI, feature extraction and dimensionality reduction is per- formed based on statistics measures, texture, local binary pattern, among others. A model based in Support Vector Regression (SVR) is trained to estimate the intramuscular fat percentage. A database of ultrasound im- ages acquired by an beef industry expert is used; for each animal there are available the intramuscular fat estimation obtained by an expert using a commercial software, and by chemical analysis. The proposed framework shows good results for a fully automatic procedure. Keywords: ultrasound images, feature extraction, intramuscular fat es- timation, beef quality, support vector regression. 1 Introduction Beef quality is a complex measure, among others, consumers highlight tender- ness as one of the most determinant factors [14,17]. It has been show that intra- muscular fat percentage (%IMF) is highly correlated with tenderness [17,2,4]. Therefore an automatic system for its measurement is fundamental. Intramuscular fat percentage is the proportion of intramuscular fat in the rib eye. This quality measure are usually performed in slaughtered animals. How- ever, is clear the importance and the utility of measuring them with the animal alive, for selective feeding, breeding, rearing [10]. For that reason is becoming important to develop automatic measurements and analysis algorithms on ul- trasound images in livestock. Ultrasound has been used in predicting beef quality for decades, allowing to measure animals’ characteristics in a non invasive way and reaching objective measures [5]. It is simple and allows real time evaluations, easy to use in a large group of animals with reasonable costs and offers an alternative for data collection in progeny testing programs. There are several previous work in this kind of applications, such as [6] ad- dressing the estimation of the %IMF in ultrasound images for livestock. In [3] the rib area was used as a determinant factor in the estimation of beef quality. M. Comin et al. (Eds.): PRIB 2014, LNBI 8626, pp. 116–122, 2014. c  Springer International Publishing Switzerland 2014