This paper presents the model and the implementation of an object classification system that with the difference to the conventional systems uses images coming from a lossy compress/decompress process. The domain transformation is made with Daubechies D4 wavelet transform and the coefficients coding is made using the Embedded Zerotree Wavelet (EZW) algorithm. The proposed model offers as main advantages a saving of until 50 % of the image total storage space and even with the information loss the images are visually very similar to the originals. The system allows obtaining very near classification results to those results obtained using the original images. In order to test the model we present the problem of apple classification. Apples are classified in two categories: bad and good quality according to specified criteria which are evaluated with a process of extraction and selection of features and the use of a voting algorithm. The results presented demonstrated that the model allows good classification results with the so important advantage that represents the storage space saving.