In circular full-cover cone-beam computed tomography (CT), the field-of-view (FOV) is limited by the width of planar detector, resulting in low imaging efficiency for large object. The FOV can be doubled by half-cover scanning, in which the back-projection filtration (BPF) algorithm based on the concept of PI-line is the best choice for image reconstruction. However, the integral intervals of different PI-lines are unequal in the BPF algorithm, leading to heavy communication consuming and calculation. As a result, the reconstruction efficiency by use of the BPF algorithm is low. In this paper, an efficient image reconstruction strategy based on the BPF algorithm for flat object is proposed. With the method, we demonstrate that the inequality of integral interval of PI-line can be ignored in the discrete implementation of the BPF algorithm when the thickness of flat object is less than 2Rsin(2π/Np) (R is the scanning radius and Np is the number of uniform sampled projections in a full circle). Compared with the original BPF algorithm for half-cover scanning, our method has two major advantages: the first one is that the outer loop is the sample angle while the inner loop is the PI-line, which reduces the communication consuming for computer significantly; the second one is that the derivative of projection, back-projection and inverse Hilbert transform along the PI-line can be computed using parallel computing techniques readily. The results of numerical simulation and real data experiment indicate that the computational efficiency of the proposed method is 5.6 times that for original BPF algorithm and the reconstruction errors of the two methods are comparable.