Performance Analysis of Gradient-Based Focus Measures in a Parallax Affected SFF Scenario

Abstract

Abstract
For scene reconstruction, shape from focus (SFF) is a popular technique in the field of computer vision. This is the fact that the SFF technique is based the focus levels of the pixels of the image preserves depth information. Conventionally, when there is a relative motion between the camera and the scene, SFF is implemented by using telecentric lenses, which avoid parallax effects. This becomes the chief component for the limitation of the SFF technique to very small objects generally in the millimeter scale. In the current research work, a new SFF-inspired algorithm is developed that uses in a wide angle lens in place of a telecentric lens. This extends the range of object that the system can deal with, though severe magnification changes occur when a stack of images are acquired with respect to the scene. By using a variable window approach the problem is addressed, when focus measures are computed. The paper presents significant results of performance evaluation of five different focus measures based on the first order image derivative, the image gradient. The evaluation is carried out based on two different performance evaluation criteria namely root mean square error and computation time. Under various operating conditions such as different spatial resolution, window size, contrast changes, gray level saturation and camera noise, the analysis of the gradient-based measures are carried.