Estimation of motion intention of human hand has many applications in robotics and other human centred areas. Especially with wearable robotic applications biosignal based estimation of human hand motions are widely used. However, based on the construction of the muscles for finger motions, and the higher number of independent motions of the human hand, estimation of finger motions accurately and effectively remains a challenge with current techniques. On the other hand, high density electromyography (HDEMG), has the capability to provide a high resolution spatial activation image of the muscle group under its measurement. In this study HDEMG signals were used to estimate the finger motions, by using the spatial variations of the surface HDEMG signals during the different finger motions. Thus, features of Gabor filters and error-correcting output codes method was used to classify six motion classes of finger motions. Results showed the proposed methodology can successfully classify the motions with a higher accuracy, using the spatial information contained in HDEMG data.