Object Based Image Analysis using Matlab

Object-based image analysis is the processing of an image based on the classification of its pixels to get useful information based on the objects contained in the image. For example, such data can be based on height, object edges, or object boundaries. <!--more--> Matlab provides an interactive environment for object-based image analysis by executing functions used in object base analysis or inbuilt apps for image processing. Images contain objects with distinct regions. These objects have boundaries, shapes, and edges.

Matlab allows for an analysis of these properties using image analyzer functions or region props to obtain data from these images. This article will discuss various edge detection methods, boundary detection, labeling of image objects, and highlighting text objects in an image.

Object-based image analysis is useful, especially in analyzing satellite maps, machine vision, fingerprint identification, and obtaining information based on object characteristics in an image.

Table of contents #

• Prerequisites
• Edge detection
• Methods of edge detection
• Sobel method of edge detection
• Prewitt method of edge detection
• Log method of edge detection
• Canny method of edge detection
• Object boundary detection
• Boundary detection using morphological operations
• Obtaining boundary by erosion
• Obtaining boundary by dilation
• Highlighting text in an image
• Labeling objects in images
• Conclusion

Prerequisites #

To follow along with this tutorial, you'll need:

• MATLAB installed.
• Basic understanding of MATLAB basics.

Edge detection #

Edge detection is the identification of points within an image. These points are where the image has excellent contrast, and they are the defining points of the image. Edge detection works by detecting changes in brightness of the image pixels.

Edge detection is useful in image segmentation and data extraction for comparison, objects separation, computer vision, and machine learning. Matlab supports a variety of functions that aids in edge detection.

Methods of edge detection #

There are several methods used in edge detection in images. These methods are used with the primary function edge(). Most edge detection methods will be demonstrated in the article using the same input image and then comparing the output image to find a more suitable edge detection method.

Edge detection syntax:

i=edge(grayscale image,'method')

All these methods only accept a grayscale image input; hence, converting the imported RGB image to a grayscale image is essential. RGB image to grayscale image conversion is done using the function rgb2gray().

The initial stage of all image processes is importing the image to the Matlab workspace. Importation of image is done using the function imread('imagefolderpath').

Let's look at an example:

i = imread('print.PNG'); %importing the image
figure,imshow(i) %  Displaying the imported image

In the above code snippet, , we import the image using the function imread().

The imported RGB image is then converted to a grayscale image using the function rgb2gray(). This is important since the methods of edge detection we will discuss only accept a grayscale image as the input image. The codes below show the conversion of RGB image to grayscale.

i2 = rgb2gray(i) %coverting the imported RGB image to grayscale image
figure,imshow(i2)% Displaying the grayscale image 

We will use this grayscale image i2 as the input image to demonstrate the following edge detection methods:

• Sobel method of edge detection.
• Prewitt method of edge detection.
• Canny method of edge detection.
• Log method of edge detection.

Sobel method of edge detection #

In the sobel method of edge detection, the edges are determined from the points with the highest gradient.

This method is executed using the function sobel() and the syntax is as shown below:

i_edge = edge(i_gray,'sobel')

To detect edges using the sobel method, we use the function edge(). Then we specify the grayscale image followed by the method used for edge detection as parameters. For example, in the code snippet below, our grayscale image is i2, then the method is sobel.

Thus, when the code is executed, the output figure assigned variable s consists of object edges detected using the sobel method.

s = edge(i2,'sobel'); % edge detection using sobel method
figure,imshow(s) %displaying edges detected using sobel method

Prewitt method of edge detection #

This is a suitable method of estimating the magnitude and orientation of edges in an image. It gives a more detailed output compared to the sobel method. The output image may be noisy depending on the number of edges present and their proximity. This method is executed using the function prewitt() and the syntax is shown below:

i_edge = edge(i_gray,'prewitt')

The following codes demonstrate edge detection using the prewitt method. The resultant image showing the detected is labeled p, the grayscale image is i2, and the method is specified as prewitt.

p = edge(i2,'prewitt'); %edge detection using prewitt method
figure, imshow(p) %displaying edges detected using prewitt method

Log method of edge detection #

This method smoothens the image then executes the laplacian function resulting in a double-edged image. This method uses the function log() and the syntax is:

  i_edge = edge(i_gray,'log')

Example:

l = edge(i2,'log'); %edge detection using log method
figure, imshow(l) %displaying edges detected using log method

The above codes implments edge detection using the log method, the resultant image containing edges detected is assigned a variable l. The grayscale input image is labeled i2 and the edge detection method is specified as log.

Canny method of edge detection #

This method detects the edges by separating noise from the image. This is better because it does not disturb the features of the edges in the image. This method uses the function canny() and the syntax is as shown below:

i_edges = edge(i_gray,'canny')

The following code snippets demonstrate edge detection using canny method, the output image is labeled c, the input grayscale image is i2 and the edge detection method is specified as canny.

c = edge(i2,'canny'); %edge detection using canny method
figure, imshow(c) %displaying edges detected using canny method

Object boundary detection #

Boundaries are lines that mark the limits of an object or an area. A boundary in images can be detected by either performing morphological operations on the image or using toolbox functions.

Boundary detection using morphological operations #

Morphological operations mainly involve subtracting parts of a binarized image with only the object's boundary. In morphological operations, boundaries can be detected by either erosion or dilation of the entities whose boundary is obtained.

The following are methods used in boundary detection using morphological process:

• Morphological erosion.
• Morphological dilation.

Obtaining boundary by Morphological erosion #

In this method, some pixels from a binarized image are eroded. The eroded part is then subtracted from the main image containing the main object. The remaining part is the boundary of the object. Erosion of image is done using the function imerode() with a specified strel() length. The following codes are step by step illustrations of this method of boundary detection.

We import the input image used to perform the operations. The image is imported using the function imread() shown in the codes below.

i = imread(plate.PNG); %importing the image to workspace
figure,imshow(i) %displaying the imported image

We then convert the imported RGB image to a grayscale image using the function rgb2gray(), the output grayscale image will be assigned i_gray as shown in the below codes.

i_gray = rgb2gray(i); % converting the imported image to grayscale image
figure,imshow(i_gray) %displaying the grayscale image

The grayscale image i_gray is converted to a black and white image using a threshold value. For our case, we will use a threshold value of 250. The black and white image will be labeled i2. The following code snippet demonstrates this process.

i2 = i_gray< 250; %making a black and white image from the grayscale image
figure, imshow (i2) %displaying the black and white image

We then specify the strel length used to perform morphological erosion. We will use a disk length of 80 to erode the black and white image i2. The eroded image will be labeled i_eroded. The below codes shows this process.

se1 = strel('disk',80); % specifying strel length for erosion process
i_eroded = imerode(i2,se1); % erosion of the image at a specified strel length
figure, imshow(i_eroded) % Displaying the eroded image

The boundary is then obtained by subtracting the eroded part of the image i_eroded from the black and white image i2 as shown in the following codes.

i_boundary = i2-i_eroded; %subtracting the eroded part of the image
figure , imshow(i_boundary) % displaying the boundary obtained

Obtaining boundary by dilation #

Dilation increases the size of the object boundary by adding some pixels to the boundary of the image. The number of pixels added depends on the size and the shape of the object being processed.

When the binarized version of the original image is subtracted from the dilated image, the resultant image is the boundary of the targeted object. Dilation of the image is done using the function imdilate() with a specified strel() length on a binary image.

The output image depends on the original image's contrast and sharpness of pixels. The following are the step by step codes used in this method of boundary detection:

The input image is imported in the function imread() shown in the codes below. The imported image is then assigned a variable i.

i = imread('plate.PNG'); %importing an image
figure,imshow(i) %Displaying the imported image

We then convert the imported RGB image i to a grayscale image i2 using the function rgb2gray() as shown below.

i2 = rgb2gray(i); %converting the imported RGB image to a grayscale image
figure,imshow(i2) %Displaying the grayscale image

The grayscale image i2 is then converted to a binary image bw1 using the function imbinarize() shown in the following codes.

bw1 = imbinarize(i2); %binarizing the grayscale image
figure,imshow(bw1) %Displaying the binarized image

We then specify the strel length used to perform morphological dilation. A disk length of 5 is then used to dilate the binary image bw1. The dilated image will be assigned i3 as demonstrated in the below codes.

se = strel('disk',5); % specifying strel length
i3 = imdilate(bw1,se); % dilation of the image at a specified strel length
figure, imshow(i3) %Displaying the dilated image

The boundary is then obtained by subtracting the binarized image bw1 from the dilated image i3 as shown in the following codes. i_boundary is assigned label for the image showing the detected boundary of the object.

i_boundary = i3 - bw1; % subtracting some parts of the image
figure,imshow(i4) %Displaying the boundaries obtained 

Highlighting text in an image #

In an image containing text messages with different words, the words are the objects contained in the image. Matlab provides functions for highlighting specified words in a text image.

We will demonstrate by highlighting the word 'MATLAB' in an image containing random letters. locatetext() function is used to locate the text to be highlighted.

The following codes are used in the process; We need to import the image containing different words using the imread() function. The imported image is assigned variable i as shown in the following code snippet:

i = imread('C:/Users/user/Pictures/TEXT.PNG'); % importing the image
figure,imshow(i) %Displaying the imported image

To view the detected letters in the imported image i, we use the function ocr(). The image text data will be assigned ocrOutput as demonstrated in the snippet below:

ocrOutput = ocr(i) %text data

Location of the words is done using the function locatetext() within the detected text data. For our case, the condition is ignoreCase, meaning the specified words will be seen and highlighted regardless of the case.

We then assign i2 to the image with highlighted words. Highlighting the words is done using the function insertshape(). Then we will use a FilledRectangle to highlight the specified words as shown in the below codes.

text_location = locateText(ocrOutput, 'Matlab', 'IgnoreCase', true); %specifying text to located
i2 = insertShape(i, 'FilledRectangle', text_location); % highlighting the located text
figure,imshow(i2) %Displaying the highlighted text in the image

Labeling objects in images #

Objects containing an image can be detected and labeled using the bwlabel() function. This function is executed in a binarized image. Furthermore, one can view the labeled object by converting bwlabel() results to RGB images using the function label2rgb().

We first import the image using the function imread(). The imported image is then assigned to variable i.

i = imread('capture.PNG'); %importing the image
figure,imshow(i) %Displaying the imported image

We then convert the imported RGB image i to a grayscale image i2 as shown below.

i2 = rgb2gray(i); % converting the imported image to grayscale
figure,imshow(i2)%Displaying the grayscale image

The grayscale image i2 is then converted to the binarized image i3 to allow the labeling of the object as shown below:

i3 = imbinarize(i2); % binarizing the grayscale image
figure,imshow(i3) %Displaying the binarized image

Labeling the objects is done by executing the function bwlabel() on a binarized image i3. The labeled objects can be viewed by converting the bwlabel() output assigned i3_label to RGB image assigned i4 using the function label2rgb(). The following codes are used in this process.

i3_label = bwlabel(i3) % labeling the objects in the binarized image
i4 = label2rgb(i3_label); % converting the labeled objects to RGB image
figure,imshow(i4) %Displaying the labeled objects

Conclusion #

Object-based image analysis helps distinguish objects contained in an image from the background part of the image by extracting certain features like boundaries and edges. Object-based image analysis has a wide range of areas of application.

It can be applied in areas like:

• Machine vision.
• Fingerprint matching for verification.
• medical diagnosis.
• Map analysis.

Happy coding!

---

Peer Review Contributions by: Miller Juma