EECS 490: Digital Image Processing

Fall 2007, Prof. Frank Merat

Student Papers Fall 2007

Identification and Tabulation of Axons using Image Processing (404kb)

Robert Alwood

ABSTRACT
This paper presents a method for the automated identification of axons using image processing. Several methods of pre-processing are used in conjunction with watershed image seg-mentation to estimate and tally the axons. Edge detection, mor-phological processing, and image thresholding were used to isolate the axons in each image before segmentation. Each method was tuned for different image types; a voting system is used to provide consistent results for a wide variety of images.

KEYWORDS
Image segmentation, watershed transform, object detection

Automated Counting Axons in Optic Nerves using Hough Transformation (135kb)

Van Anh Tran

ABSTRACT
I developed an algorithm to count the number of axons in optic nerves automatically using Hough Transformation to detect axon edges. Due to the low quality of nerve images, a pre-process is required using Gaussian filter.

KEYWORDS
automated counting, axons,optic nerve, Hough transformation

Automated Axon Counting via Digital Image Processing Techniques in Matlab (328kb)

ABSTRACT
This paper presents the design of an algorithm that uses various image enhancement and image processing techniques to count axons in the optical nerve of a mouse. The digital image is the cross section of the nerve stemming from the eye in route to the brain. The goal of the algorithm is to be able to accurately detect the axons and give the user a count of the number axons present in the image.

KEYWORDS
Cell segmentation, image segmentation, axons, image Processing, adaptive thresholding, watershed, Matlab, morphological

Automatic Quantification of the Axons in the Optical Nerve of a Mouse (528kb)

ABSTRACT
This paper presents a morphological method of automatically identifying and quantifying the nerve bundles in a digital image of the optical nerve of a mouse. Mice are commonly used in biological and genetic experiments. It is often necessary to count the rods and cones in the retina. Because of the curvature of the retina it is often easier to count the axons in the optical nerve. Automating this process of quantification is the goal of this paper.

KEYWORDS
Segmentation, optic nerve, mouse, retina, ganglion, image morphology, automatic quantification

Image Resizing by Seam Carving in Python and Matched Masks (2.2Mb)

ABSTRACT
This paper explores a recently developed technique called “seam carving” [1] to remove low energy seams from the image to create a crop that preserves more information. Rather than interpolating new pixels or removing a rigid column of pixels, seam carving removes fluid seams of 8-connected pixels. It further explores the concept of a matched mask to prevent distortion.

KEYWORDS
Image, resize, crop, retarget, seam carving, retargeting, matched mask, Python

Algorithm for Detecting and Counting Tightly Bundled Axons in Normal Optic Nerve (1.1Mb)

ABSTRACT
This paper demonstrates a way to effectively identify and count cross-sectioned Axons bundled in a tightly packed optic nerve. The main challenge in this application is the extremely high cellular density and low contrast. The method described in this paper uses local adaptive thresholding followed by various morphological operations to segment individual cells. The segmented cell candidates are then evaluated for specific physical characteristics and the original image is augmented with an overlay showing counted cells and rejected cell candidates. The average accuracy among all 6 test scenarios was 87% correct identifications and having an overall count 6% below counts performed by a trained researcher.

KEYWORDS
Adaptive filtering, image processing, edge detection, biological, axon, segmentation, morphological operations.

Algorithm for Identifying Axons (248kb)

ABSTRACT
This paper presents a method for detecting and counting axons, or nerve fibers, in an image. This method is demonstrated on images containing a cut-away view of a nerve bundle, which contains many axons that must be identified and counted. This technique can be applied to a variety of applications of similar nature. The method utilizes spatial filtering, morphological operations, threshold, and binary image manipulation techniques.

KEYWORDS
Axon identification, histogram equalization, morphological operations, threshold

Automated Biological Image Segmentation (292kb)

ABSTRACT
This paper presents a method for automated identification and counting of neural connections (axons) in a microscope image of an optical nerve bundle. The algorithm uses morphological image processing techniques, coupled with traditional image segmentation methods, to generate a binary image suitable for counting features. Results from goldstandard images show the algorithm accuracy to be highly dependent on both the average size of axons and image quality (i.e. sharpness, contrast and resolution). Input images containing several axons of similar size show the algorithm to be reasonably accurate and robust. The algorithm breaks-down when processing grainy, pixilated nerve bundles containing few, large axons.

KEYWORDS
Biological image segmentation, retinal sensing, optical nerve bundle, axon counting, morphological image processing

Hybrid Method of Biomedical Image Segmentation (536kb)

ABSTRACT
In this paper we present a general frame work for axon segmentation in retinal image and count the number of axons. The algorithm contains three phases. First phase is preprocessing. The goal of this phase is to get a binary image with cell objects in it. In this section, we use two different thresholding methods. One is FCM clustering and the other is Otsu's method and I also compare the results of both methods. The next phase is separate cell objects from the binary image after phase 1. Here we separate the objects based on its size and shape. To separate objects by shape we use V/H scan line shape detection to check the smoothness of objects. The last phase is to count the number of object in a binary image.

KEYWORDS

Counting Mouse Neuron Cells Using Morphological Image Processing (504kb)

ABSTRACT
Various methods exist for counting cells in medical images. Simple statistical methods exist, such as the Abercrombie and Empirical methods,which simply use small sections of a set of images to estimate a full count. These methods have flaws due to the large number of assumptions made with them (Hedreen, John C, 1998). Stereoscopic and 3D methods also exist, but require some kind of 3D information in order to be effective. In this paper, I have designed an algorithm that automatically counts the number of cells in a single, 2D image, with only minor thresholding adjustments needed depending on cell sizes and image gamma.

KEYWORDS

Biological Image Segmentation (860kb)

ABSTRACT
This paper presents an approach of using multiple image processing techniques including color segmentation, image smoothing, contrast enhancing, edge processing, threshold-ing and morphological reconstruction processing to iden-tify and count the number of rods and cones in a mouse retina microscope image. Different approaches of image preprocessing have also been tested and the results have also been compared to find an optimal solution for this purpose. An adaptive background area removing method has also been addressed in this paper. This automated counting and adaptive background eliminating algorithm allow the researches to separate the target objects from the background and reduce the rate of overcounting.

KEYWORDS
Image analysis, image segmentation, segmentation evaluation, closed bundles detection

Identification and Counting of Mouse Axons in a Retinal Scan (40kb)

ABSTRACT
This paper presents a method for detecting and counting the number of axons present on a retinal scan from a mouse. The algorithm used incorporates morphological transformations, including top hat, fill, and shrink transformations. The results from test cases indicate that the algorithm does indeed identify all of the axons, but has trouble filtering out the blank space in between axons.

KEYWORDS
Image processing, feature detection, retina axons

Cells Segmentation Using the Hybrid of Image Morphology and Edge Detector Algorithm and Cell Counting (804kb)

ABSTRACT
In recent decades, demand for applying image-processing techniques of improving biomedical diagnosis has been raising rapidly because traditional system process does not only cost too much time and human resources but also causes a lot of errors and unexpected results. In addition, traditional method requires skilled experts to work on it so that only minority can be qualified. As an objective and programmed method is required to eliminate those shortcomings discussed above. In this paper, I propose the hybrid of Image Morphology and Canny Edge detector Algorithm (HIMCA) to segment, count, and locate sensors in retina automatically. In addition, I also offer a feedback control scheme to correct the errors HIMCA makes. HIMCA can improve the performance of cell segmentation a lot although there are still some cells hard to tell. However, it has a good chance to solve this problem in the near future.

KEYWORDS
Cell counting, image morphology, morphology, edge detector, segmentation.

Automated Axon Counting in Normal Optic Nerves Using Image Morphology and Area Based Segmentation (208kb)

ABSTRACT
This paper presents an application of image morphology and segmentation in the field of medical imaging for the use of ascertaining the number of axons from cross section images of an optic nerve. Because of the complexity of the nerve and inconsistency of individual axon sizes and shapes, a general approach is conceived to provide an accurate estimation of the number of axons present. By taking into account image statistics and evaluating regions according to their characteristics, results can be obtained that count axons to within a few percent of the actual number present.

KEYWORDS
Axon counting, normal optic nerve, image processing, computer vision, segmentation, morphology

Autonomous Universal Optic Nerve Axon Delineation from High Resolution Digitized Samples (672kb)

ABSTRACT
This paper presents a universal algorithm designed to delineate and count axioms precisely from digitized optic nerve cross-sections. The algorithm is designed to be used with any sized image and segments the work so it can be executed simultaneously on multiple computers and/or processors. With the application of filtering, histogram adjustments, edge detection, and size masking we are able to detect cellular structures in the image precisely while removing staining artifacts. A simple methodology is demonstrated in counting cells distorted with myelinstained irregularities and its application to multiple samples in a production environment.

KEYWORDS
Optic nerve, imaging, canny, Otsu, automated, axon

Counting Axons in an Optical Nerve Bundle of a Mouse Using the Digital Image Processing Techniques (168kb)

ABSTRACT
This paper presents an investigation on using standard digital image processing techniques to quantify the number of axons in an optical nerve bundle of a mouse. The standard methods of processing will include edge enhancement, morphology and thresholding. The inherent limitations of the above methods will be delineated and an alternative method developed. The alternative method demonstrated will be based on statistical methods along with an initial calibration process. Once an image of particular process is calibrated the number of axons can be counted for any nerve bundle image that has been produced from this process. The scope of the defined alternative process will be limited to counting axons without regard to their location in the image.

KEYWORDS
Segmentation, morphology, axon, edge detection

Counting the Number of Axons from Mice Optical Bundles using Morphological Image Segmentation (360kb)

ABSTRACT
This paper presents the design of an application with digital image processing algorithms in MATLAB. This algorithm allows for this application to count the number of axons of the optical bundle from mice, which were obtained from an image. The images were supplied by Prof. Howell at Case Western University who is researching retinal sensing. He is looking for a way that automates the location and counting of closed bundles of axons from the supplied images. This application attempts to do that with good precision. This GUI application allows a user to open and display the image that will be counted. The application uses default values to determine the best threshold for processing the image (threshold is also user selectable). A second image then displays the found axons, and a text field displays the numeric count and processing times.

KEYWORDS
Counting axons, morphological image segmentation, MATLAB, auto threshold

The Use of Image Segmentation in the Detection of Rods and Cones in the Retina of Mice (440kb)

ABSTRACT
Six images of rods and cones in the retina of mice were provided. It was desired that these rods and cones be automatically counted. Several different image processing methods were used to count the number of rods and cones in each image. These methods included adaptive histogram equalization, edge detection, and morphological operations. This paper describes the algorithm used to count the rods and cones in the retina of mice.

KEYWORDS
adaptive histogram equalization, edge detection, image processing, thresholding, contrast adjustment, image segmentation

Robust Automated Algorithm for Counting Mouse Axions (716kb)

ABSTRACT
This paper presents a robust technique for counting axion cells within a digital image of a mouse neuron bundle. The method involves employing Matlab to transform the color image into grayscale, performing regional histogram equalization, thresholding the image, and finally detecting cell boundary using morphological technique. This algorithm detects and counts cell boundaries with 92% accuracy.

KEYWORDS
Mouse Neurons, cell counting, computer vision, background filtering, regional histogram equalization, thresholding, image segmentation, morphology, boundary detection.

A Hole-filling Algorithm for Automated Axon Counting (120kb)

ABSTRACT
This paper presents an algorithm for the automated counting of axons in microscope images of cross-sections of the optic nerve. The approach uses a hole-filling algorithm to identify closed bundles in the image. Six images in which axons were labeled by a trained observer (TO) were used as test data for the algorithm. Results show that the algorithm counted the axons with an average relative error (n=5) of (34+/-11) %. This error is far too large for the algorithm to be useful at present. It is believed however that this error is due primarily to the image segmentation method used, and that the algorithm could be improved with some modifications to increase performance.

KEYWORDS
Automated counting, optic nerve, imaging/image processing/image analysis

Axon Detection in Digital Images (236kb)

Paul C. Whitten

ABSTRACT
This paper presents two algorithms used to successfully identify cellular structures in digital images. More specifically, given digital images of stained and magnified axons from a cross section of an optic nerve, this work applies various digital image processing techniques in order to detect and count the axons. The techniques explored in this paper are Canny edge detection, thresholding (optimum, and hysteresis), morphological pruning, and the Moore boundary tracking algorithm. Algorithms composed of combinations of these techniques are presented along with additional steps used to improve results. Finally, the results obtained are compared to those provided by trained professionals.

KEYWORDS
Digital image processing, edge detection, Canny, cellular structres

Location of Closed Bundles in an Optical Nerve Using Boundary Tracking of a Digital Image (352kb)

David Young

ABSTRACT
This paper presents an application of image segmentation to locate and count individual nerve bundles from the optical nerve of mice. Automated image processing techniques such as color-space manipulation, histogram analysis, and boundary definition are used to reach the solution. To further increase the accuracy of the algorithm, minimal input is required from the user to allow the program to automatically set parameters.

KEYWORDS
Image processing, image segmentation, histogram, binary image, nerve bundles