2
$\begingroup$

i have to do super resolution of two low resolution images to obtain a high resolution image.

1st image is taken as base image and the second image is registered with respect to it . i used SURF algorithm for image registration . A Delaunay triangulation is constructed over the points using a built-in MATLAB delaunay function . The HR grid of size is constructed for a prespecified resolution enhancement factor R Then HR algorithm for interpolating the pixel values on the HR grid is summarized next.

HR Algorithm Steps: 1. Construct the Delaunay triangulation over the set of scattered vertices in the irregularly sampled raster formed from the LR frames.

Estimate the gradient vector at each vertex of the triangulation by calculating the unit normal vector of neighbouring vector using cross product method.Sum of the unit normal vector of each triangle multiplied by its area is divided by summation of area of all neighbouring triangles to get the vertex normal.

Approximate each triangle patch in the triangulation by a continuous and, possibly, a continuously differentiable surface, subject to some smoothness constraint. Bivariate polynomials or splines could be the approximants as explained below. Set the resolution enhancement factor along the horizontal and vertical directions and then calculate the pixel value at each regularly spaced HR grid point to construct the initial HR image

now i have the results shown below

now for one kind of data set i get this result that has a few pixels black and wite in a random manner for the other type i get thin parallel lines all over image after super resolution the results are attached

for first set random pixels appear on super resolution,for second set random pixels with parallel lines appear, these pixels appear on differnt pattern when i change transformation type from affine to projective in estimate geometric transform command, more over for other data sets there are deformation like parallel lines in edges.for some images it is quite well and does not give any deformation

any one can tell me the reason, i have figured out may be its demosaicing but i am not sure,because i dnt have much understanding of it , moreover can it be a bug in my code but it behaves different for different images, i have increased the size by super resolution twice. superresolved image first unregistered image unregistered image second unregistered image unregistered imagesecond unregistered image

$\endgroup$
  • $\begingroup$ any suggestions please? i figures out image interlacing as well $\endgroup$ – Jav Jan 29 '14 at 13:37
  • $\begingroup$ Off-the-cuff guess: truncation error from float to integer. Impossible for me to say without seeing code. $\endgroup$ – MackTuesday Nov 5 '15 at 20:25
  • $\begingroup$ I am working on the same problem. And I followed the steps given in the paper but I am getting gray images as a result. Could you help me understand what could be the possible reason for that, since you have already worked on it? $\endgroup$ – user19861 Mar 3 '16 at 19:00
  • $\begingroup$ Could you share the code and the paper you used? We can't help you like that. $\endgroup$ – Royi Mar 3 '16 at 20:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.