Focused-Region Segmentation for Refocusing Images from Light Fields
Received: 21 March 2018 /Revised: 2 May 2018 /Accepted: 8 May 2018 / Published online: 27 May 2018
Springer Science+Business Media, LLC, part of Springer Nature 2018
Since the focused regions of the light-field refocusing image contain depth cues, focused-region segmentation becomes a
fundamental step of depth estimation, 3D measurement, and visual measurement. However, in the emerging field of light-field
image processing, recent research has emphasized focus evaluation rather than a systematic method for focused-region segmen-
tation. The segmentation algorithms for low depth-of-field images are of significance for this problem, but those algorithms have
high time complexity and are not suitable for the computationally intensive applications of light-field imaging. Therefore, based
on the pulse synchronous mechanism of the pulse coupled neural network (PCNN), we establish the model of neural firing
sequence and some criteria of pixel classification. Further, we design an algorithm of the focused-region segmentation and its
parameter settings. The experimental results show that the proposed method segments the refocusing images faster than alter-
native methods and meets the needs of light-field image processing and related applications.
Keywords Light field
Pulse-coupled neural networks (PCNN)
In recent years, microlens light-field cameras have emerged
not only in the consumer market but also in the industrial
market, providing new solutions for application areas such
as industrial inspection, visual measurement, 3D reconstruc-
tion and depth estimation [1–4]. Unlike conventional cameras,
microlens light-field cameras record the location and direction
of light rays simultaneously. Three-dimensional measurement
using microlens light-field cameras has unique advantages.
First, an unfocused plane can be refocused after exposure.
Second, the aperture size is arbitrary as long as it does not
exceed the synthetic aperture, with a depth-of-field that is
adjustable through calculation. Third, light-field images can
be processed and used directly in 3D measurement applica-
tions [5, 6].
The key step in the application of light-field data to 3D
measurement is the detection and segmentation of the focused
region of the light-field refocusing image. Within some geo-
metrical constraints, the focused region of a light-field image
contains distance and depth information. However, light-field
image processing is still an emerging field and lacks compre-
hensive research. To date, the emphasis of research has been
focused object segmentation algorithms for images with low
depth-of-field. However, these algorithms prioritize the seg-
mentation of the complete object, resulting in a high compu-
tational complexity that makes them unsuitable for applica-
tions such as 3D measurement. In this paper, we propose a
focused-region segmentation method based on an improved
pulse-coupled neural network (PCNN) . This method seg-
ments the refocusing image and achieves the two-fold goal of
accuracy and low computation time.
The contributions of this paper can be briefly summarized
1) The model of the neural firing sequence is built on the
pulse output of PCNN, and some criteria is presented to
divided the pixels into different classifications.
2) A focused-region segmentation algorithm based on
PCNN is proposed to segment focused-region from light
3) An automatic parameter setting method is designed using
the statistical features of the input image.
* Jie Shen
College of Computer and Information Engineering, Hohai
University, Nanjing 211100, China
School of Computer Engineering, Nanjing Institute of Technology,
Nanjing 211167, China
Journal of Signal Processing Systems (2018) 90:1281–1293