Three-dimensional shape recovery from the focused-image surface

3D shape recovery from image focus using kernel regression in eigenspace

“…In this method, a rough estimate is taken by applying F SML and then the best-focused values lying on FIS are searched in second step. Choi and Yun [19] proposed an estimation of FIS through piecewise curved surface approximated through interpolation by using second order Lagrange polynomial. Asif and Choi [20] applied neural network to optimize the FIS.…”

“…Thus, loss of focusing information between two connective frames results in inaccurate depth map. In literature, curve fitting and surface approximation techniques have been suggested to overcome this limitation [6,19]. These techniques are the special cases of parametric regression, and are based on assumption that the focus val- ues follow some specific distribution models.…”

Optimal depth estimation by combining focus measures using genetic programming

“…However, this method is computationally expensive as it searches the plane that provides optimal focus measure from a huge number of possible planes. Further extending their work, Choi and Yun [7] suggested the estimation of FIS through piecewise curved surface approximation instead of planar window as the FIS of general objects is usually not planar. They used the Lagrangian polynomials with nine control points to approximate the FIS in a window around each pixel.…”

A Fuzzy-Neural approach for estimation of depth map using focus

“…For example, Choi and Yun [11] proposed the approximation of focused image surface by a piecewise curved surface rather than through the use of a piecewise planar approximation, where the piecewise curved surface is estimated by interpolation using a second order Lagrange Polynomial. And Asif and Choi [12] used Neural Networks on GLV result to learn the shape of focused image surface by optimizing the focus measure over small 3D windows, as due to their nonlinear characteristics, Neural Networks can be used to approximate any arbitrary function.…”

“…The backward difference of focus measure will be positive and forward differ- ence will be negative around peak focus measure computed by using Sum of Modified Laplacian (SML) as the focus measure. We selected to use SML focus measure because this is the most widely reported as well as used focus measure for implementation of various approximation methods [6,[9][10][11][12][13] explained in Section 2.2. The input variables, backward difference and forward difference of peak focus measure, are modeled in fuzzy logic as shown in Fig.…”