Electrical resistance tomography is a soft-field tomography technique, i.e. the electrical field is changed everywhere in the sensing area with the change of conductivity in any pixel. To improve the image quality, an extended sensitivity matrix is designed in this paper. The base conductivity elements in the extended sensitivity matrix are consisted of a series of blocks with different number of pixels at all possible locations in the sensing region. Based on the new sensitivity matrix, a sparse representation method is implemented to reconstruct the conductivity distribution of cross-sectional area. Simulation results show that the proposed method based on the extended sensitivity matrix can reconstruct the image with a high quality.