Chapter 6. Learning Object Classification

In the previous chapter, we introduced you to the basic concepts of object segmentation and detection. This means isolating the objects that appear in an image for future processing and analysis.

This chapter covers how to classify each of these isolated objects. In order to allow us to classify each object, we need to train our system to be capable of learning the required parameters to decide which specific label should be assigned to the detected object (depending on the different categories taken into account during the training phase).

This chapter is going to introduce you to the basic concepts of machine learning to classify images with different labels.

We will create a basic application based on the segmentation algorithm, as discussed in Chapter 5, Automated Optical Inspection, Object Segmentation, and Detection. This segmentation algorithm extracts parts of an image, which contains objects. For each object, we will extract the different features...

Introducing machine learning concepts

Machine learning is an old concept that was defined in 1959 by Arthur Samuel as a field of study that gives computers the ability to learn without being explicitly programmed. Tom. M. Mitchel provided a more formal definition. In this definition, Tom links the concept of samples or experiences, labels, and performance measurements.

Note

The machine learning definition by Arthur Samuel is referenced from Some Studies in Machine Learning Using the Game of Checkers in the IBM Journal of Research and Development (Volume: 3, Issue: 3), p. 210 and a phrase in The New Yorker and Office Management the same year.

The more formal definition by Tom. M. Mitchel is referenced from Machine Learning Book, McGray Hill 1997 (http://www.cs.cmu.edu/afs/cs.cmu.edu/user/mitchell/ftp/mlbook.html).

Machine learning involves pattern recognition and the learning theory in artificial intelligence and is related to computational statistics.

Machine learning is used in hundreds of applications...

Computer Vision and the machine learning workflow

The Computer Vision applications with machine learning have a common basic structure. This structure is divided into different steps that are repeated in almost all Computer Vision applications, and some others are omitted. In the following diagram, we show you the different steps involved:

Almost any Computer Vision application starts with a preprocessing stage that is applied to the input image. Preprocessing involves light removal conditions and noise, thresholding, blur, and so on.

After we apply all the preprocessing steps required to the input image, the second step is segmentation. In the segmentation step, we need to extract the regions of interest of an image and isolate each one as a unique object of interest. For example, in a face detection system, we need to separate the faces from the rest of the parts in the scene.

After getting the objects inside the image, we continue with the next step. We need to extract all the features of...

Automatic object inspection classification example

Continuing with the example of the previous chapter, the automatic object inspection segmentation, where a carrier tape contains three different types of objects (nuts, screws, and rings), and with Computer Vision, we will be able to recognize each one of them to send notifications to a robot or similar to put each one in different boxes.

In Chapter 5, Automated Optical Inspection, Object Segmentation, and Detection, we preprocessed the input images and extracted the regions of interest of images and isolated each object using different techniques. Now, we will apply all these concepts, as explained in previous sections, in this example to extract features and classify each object and allow to possible robot to put each one in different boxes. In our application, we are only going to show the labels of each image in an image, but we can send the positions in the image and the labels to other devices as a robot.

Then, our goal is from an input...

Feature extraction

Now, let's extract the features of each object. To understand the feature concept of a feature vector, we will extract very simple features, but it is enough to get good results. In other solutions, we can get more complex features, such as texture descriptors, contour descriptors, and so on.

In our example, we only have these three types of objects, nuts, rings, and screws, in different possible positions. All these possible objects and positions are shown in the following figure:

We will explore the good characteristics that will help the computer to identify each object. The characteristics are as follows:

• The area of an object

• The aspect ratio, which is the width divided by the height of the bounding rectangle

• The number of holes

• The number of contour sides

These characteristics can describe our objects very well, and if we use all of them, the classification error can be very small. However, in our implemented example, we will use only the first two characteristics, the...

Summary

In this chapter, we learned the basics of the machine learning model and how to apply a small sample application to understand all the basic tips required to create our own ML application.

Machine learning is complex and involves different techniques for each use case (supervised learning, unsupervised, clustering, and so on), and we learned how to create the most typical ML application and the supervised learning with an SVM.

The most important concepts in supervised machine learning are: first, we need to have an appropriate number of samples or datasets; and second, we need to correctly choose the features that describe our objects correctly. For more information on image features, refer to Chapter 8, Video Surveillance, Background Modeling, and Morphological Operations. Third, choose the best model that gives us the best predictions.

If we don't reach the correct predictions we have to check each one of these concepts to look for where the issue is.

In the next chapter, we will introduce...