Optimizing Focus Measure Approach and Enhancement of Image Focus Volume in 3-D Shape Recovery based on Shape from Focus Architecture

Abstract

Recovering an accurate and precise 3-D shape of unknown microscopic object using shape from focus architecture is an involved problem in computer vision community. A microscopic object is placed on a translational stage which is capable of moving up and down in optical axis of camera. A stack of images is gathered by capturing each image at the discrete steps. During image acquisition step, the stage is translating with respect to the camera while the RGB camera sensor is stationary at a predefined known height. After image acquisition step, a focus measure algorithm is applied on the gathered stack of images to convert the corresponding RGB color information into focus quantity at each pixel. Initial depth map is acquired by maximizing the focus measure curve along the optical axis of camera. Various noise sources, mechanical disturbance during image acquisition or inaccuracy of RGB camera mainly leads to the false depth map which greatly suffers the acquired 3-D shape. An approximation approach or a regression algorithm is mainly employed after the focus measure algorithm. Initial depth map has discreteness along each frame and have noise error. Initial depth map is further refined by applying the most recent approximation algorithms, regression techniques or the complex filters. Recovering 3-D shape using shape from focus architecture has numerous industrial and medical applications and is currently practiced in most of the developed countries