TREES

Background

 
 

Introduction

TREES is a cooperative, multidisciplinary project at the University of Arizona between the Laboratory of Tree-Ring Research and Digital Image Analysis Lab of the Electrical & Computer Engineering department.  It is funded by Grant SBR9601867 from the National Science Foundation and by the University of Arizona, Office of the Vice President for Research.

The purpose of the TREES project is to develop a computer-assisted environment for tree-ring analysis through the use of digital image processing and computer vision techniques.

Here is an example of tree ring detection.
 

Background

The science of tree ring dating, known as dendrochronology, provides techniques for the precise dating of trees.  During each year in the lifetime of many trees, a single tree ring is created.  The outer-most ring at the tree's bark corresponds to the current year of a live tree or the last year of growth of a dead tree, and by counting the rings inward toward the center of the tree it is possible to tell how many years the tree lived.  Moreover, the widths of tree rings vary from year to year, creating patterns of variation that are present across different trees in a geographical region.  These patterns make it possible to cross-date between tree samples.

Researchers invest many hours manually examining tree ring samples under a microscope. By tediously measuring tree ring features such as ring boundaries and widths and correlating data from different samples, they are able to perform dating so precise that Carbon-14 dating is calibrated from their results.  Despite advancements in computational speed and image processing algorithms, manual measurement techniques remain the primary approach for collecting tree ring data since no previously developed automated technique is reliable enough for widespread use.

The methods being developed in the TREES project provide a means for increasing the efficiency of tree ring analysis.  Rather than attempting to produce a fully automated solution, the philosophy of this project is to create a computer-assisted environment that integrates analyst intervention with algorithmic decisions.  This approach allows the development of a robust system that can emulate the complex human vision analysis of tree samples, which often tend to have unpredictable features.
 

Overview of TREES System

The computerized tree-ring dating system is being designed to resolve small tree rings while limiting the amount of memory required to image a tree sample.  Tree rings can range from 0.05mm in width to several centimeters, although most are around 1mm or less.  Typical tree sample sizes range from 500mm to over 100cm in length.  The hardware chosen for this purpose consists of a 1317x1035 Kodak Megaplus 1.4i CCD digital camera attached to a Nikon SMZ-U microscope for imaging across section or core sample of a tree situated on a positioning table under the microscope.  The microscope is generally set to a magnification of 1.1X.  In order to increase the field of view of captured frames, a video coupler lens with a magnification of 0.63X is used, creating a combined magnification of 0.693X and a diagonal field of view of approximately 16.4mm for each frame.  This results in a pixel size on the sample of approximately 10um, which is sufficient to resolve most small tree rings.  For illumination, a fiber-optic ring illuminator with a diameter of approximately 75mm is mounted on the objective lens of the microscope.  The positioning table includes two stepper motors under computer control for motion in the X- and Y-directions of up to 75cm, with a precision of about 1mm.  The table is belt driven (rather than lead screw driven) for faster positioning over many centimeters.  Software correlation techniques are applied to make up for the low precision of the positioning table when combining multiple captured frames into a single mosaiced image.  Finally, a stepper motor is also used to control the Z-axis for focus adjustments.  The camera, microscope, focus, and positioning table are all controlled by a Sun ULTRA 1 workstation with 128MB of memory, 8.2GB disk capacity, and a 4MB frame buffer.  With the current configuration, a monochrome mosaiced image of a 500mm by 5mm tree core sample can be created from 50 overlapping frames and occupies approximately 54MB.  The time required to acquire the frames and automatically create the mosaic is under ten minutes.

For the first step of the tree ring analysis process, a series of overlapping images are acquired across the tree sample, from the center of the tree (the pith) to the outer edge (the bark).  These images must be corrected for any illumination distortion and normalized for any variation in camera gain and brightness offset during capture.  The individual frames are then registered to one another through correlation techniques into a single mosaic.  It is then possible to apply edge detection techniques to find the locations of tree ring boundaries in the sample mosaic and tree ring measurements can be made.  Ring widths and other characteristics may be used to match ring patterns from one tree to another, allowing cross-dating to be performed.

Several major philosophies guide the development of this system.  First, the goal is to create a computer-assisted, not automated, tree ring dating system.  When dealing with natural phenomena such as tree rings, it is virtually impossible to predict all special cases.  Because every tree ring must be accounted for to accurately cross-date a tree sample, it is better to allow analyst intervention when difficulties arise rather than attempting to develop a complex algorithm to make decisions with a low probability of error under all conditions.  Also, to improve accuracy and reliability, the system is wood-centered.  Rather than relying solely on digital images, the analyst can visually explore the wood surface through the microscope at any point in the tree ring analysis process to resolve problems of ring identification and dating.  The system is primarily meant to increase the efficiency of tree ring analysis by detecting and measuring the majority of the tree rings in a sample with minimum operator intervention.  However, the operator may use expertise in dendrochronology for verification, to provide suggestions to the system, or to detect subtle or ``false'' rings.

At the time of writing, the computer vision problems described in this paper are only partially solved.  The algorithms and techniques included here represent the best methods found to date to solve the challenges of tree ring analysis.