VRL an eye

Multi-focus imaging

People

Dmitry V. Fedorov, Baris Sumengen, B.S. Manjunath

Objective

Modern optic systems carry several fundamental limitations. One of them is the low depth-of-field. Usually, certain objects at particular distance are focused while other objects are blurred to a degree depending on their distances from the camera (see Fig. 1). This problem is encountered in photography and microscopy.

Img. 1a) first image of 9 Img. 1b) last image Img. 1c) Combined result Img. 1d) Depth map

Our framework for image processing using local information provides a solution. Our approach is robust to acquisition parameters and temporal changes. Since we blend the images using pixel data from the spatial domain (as opposed to fusing the information in a transform domain), the resulting images have fewer artifacts. The results presented for microscopy and hand held consumer cameras demonstrate good quality and computational efficiency.

Example 0

This example shows the comparison of our result with the method from "Photomontage" paper by Aseem Agarwala. Images were retreived from the author's website

Img. 2a) 1st of input images Img. 2b) 4th of input images Img. 2b) 8th of input images Img. 2c) Our result Img. 2e) Result by the "Photomontage" method

Example 1

Two images of hydrant with focus setting at 20cm (a) and 2m (b). Consumer camera positioned on the tripod, the images present small temporal changes due to wind. c) Combined resultant image.

Img. 2a) Focus at 20cm, flowers are focused Img. 2b) Focus at 2m, hydrant is focused Img. 2c) Combined result

Example 2

a-b) Two of the four slightly misaligned images of integrated optical waveguide acquired by Scanning Electron Micrograph (SEM). c) Our result. d) The result of wavelet image fusion.

Img. 2a) Focus on etched wafer surface behind waveguide, left upper corner Img. 2b) Focus on tilted waveguide sidewall, right lower corner Img. 2c) Our multi-focus result Img. 2d) Wavelet image fusion

Example 3

Three images of fire with focus setting at 20cm (a), 2m (b) and infinity (c) from hand held consumer camera and combined resultant image (d).

Img. 2a) Focus at 20cm Img. 2b) Focus at 2m Img. 2c) Focus at infinity Img. 2d) Combined result

Example 4

Three images of monument with focus setting at 20cm (a), 2m (b) and infinity (c) from hand held consumer camera and combined resultant image (d).

Img. 2a) Focus at 20cm Img. 2b) Focus at 2m Img. 2c) Focus at infinity Img. 2d) Combined result

Example 5

Two images of plants with focus setting at 2m (a), and infinity (b) from hand held consumer camera and combined resultant image (c).

Img. 2a) Focus at 2m Img. 2b) Focus at infinity Img. 2d) Combined result

Example 6

High Resolution Through-Focus Movie of 40 frames (a,b) through the dorsal lateral geniculate nucleus and the in-focus result (c) with cell counting result in yellow. *Video provided by Robert W. Williams from Department of Anatomy and Neurobiology, University of Tennessee.

Img. 2a) Frame 10 Img. 2b) Frame 28 Img. 2d) Combined result of 40

Acknowledgements

This project is supported by the NSF Information Technology Research grant #0331697.

Publications

  1. Dmitry Fedorov, Baris Sumengen and B. S. Manjunath,
    "Multi-Focus Imaging using Local Focus Estimation and Mosaicking"
    Proc. IEEE International Conference on Image Processing 2006 (ICIP06), Atlanta, GA USA, Oct. 2006.
    [abstract] [PDF] [BibTex]

    Abstract preview: "We propose an algorithm to generate one multi-focus image from a set of images acquired at different focus settings. First images are registered to avoid misalignment if needed. Then, best focus is es..." [more]