An approach for the 3D reconstruction of architectural scenes from two uncalibrated images is described in this paper. From two views of one architectural structure, three pairs of corresponding vanishing points of three major mutual orthogonal directions can be extracted. The simple but powerful constraints of parallelism and orthogonally in architectural scenes can be used to calibrate the cameras and to recover the projection matrices for each viewpoint. The projection matrices are used to reconstruct a partial 3D model of an architectural scene from two uncalibrated photographs taken from arbitrary viewpoints. The approach is applied to the real images of architectural scenes, and a 3D model of a building in VRML format is presented which illustrates the method with successful performance. It is applied to recover 3D models using the hand held digital cameras that the camera motion can't be controlled.

In this paper we report our investigation on zonal sampling, probability fitting tests, and bit allocation for Hadamard transform image data compression. The energy distribution of the transform coefficients suggests a sampling strategy which retains as much energy as possible for a fixed number of coefficients. The probability fitting tests indicate that Laplacian is a better model for non-dc coefficients, while Rayleigh or Gaussian fits better for the dc coefficient. Finally, the bit assignment is formulated as a constrained optimization problem. The mean square error performance is derived as a function of the number of retained coefficients for various average bit rates. The results are useful for designing a zonal coding system with optimum retained zone size and shape as well as their corresponding bit assignment.