The estimation of the cavings depth is of particular interest in the oil industry. During the drilling process, the rock classification problem has been studied in order to analyze the cuttings concentration at the vibrating shale shakers, through the classification of the cavings images. Nevertheless, depth estimation based on cavings rock images has not been treated in the literature. This paper presents a new depth caving estimation system based on the classification of the caving image through feature extraction. To extract the texture descriptors, the cutting images are first mapped on a common space where they can be easily compared. Then, texture features are obtained by the application of a multi-scale and multi-orientation approach, through the use of Gabor transformations. Two different depth classifiers are developed, the first separates the texture features by using a soft decision based on the Euclidean distance, and the second performs a hard decision classification by applying a thresholding procedure. A detailed mathematical formulation of the developed classifiers is presented. The developed estimation system is verified using real data from cutting rock images in petroleum wells. Several simulations illustrate the performance of the proposed model using real images of a wellbore in a Colombian Basin. The performance of correct classification by a 17 depths estimation database is 91.2 %.