Chapter 8: High-Pass Filters and Feature Detection

In the previous chapter, we learned about kernels and low-pass filters and their applications. We learned about and demonstrated how to use low-pass filters in blurring, smoothing, and de-noising images.

In this chapter, we will learn about and demonstrate the uses of high-pass filters. This includes their application in image processing and computer vision. First, we will explore the Laplacian, Scharr, and Sobel high-pass filters. Then, we will learn about the Canny edge detection algorithm. We will also demonstrate Hough transforms for circles and lines. We will conclude by looking at corner detection with the Harris algorithm.

The following is a list of the topics we will cover in this chapter:

• Exploring high-pass filters
• Working with the Canny edge detector
• Finding circles and lines with Hough transforms
• Harris corner detection

After following this chapter, you will be able to use high-pass filters...

Technical requirements

The code files of this chapter can be found on GitHub at https://github.com/PacktPublishing/raspberry-pi-computer-vision-programming/tree/master/Chapter08/programs.

Check out the following video to see the Code in Action at https://bit.ly/2CFnpnD.

Exploring high-pass filters

The concept of high-pass filters is exactly the opposite of low-pass filters. High-pass filters allow high-frequency components of information (such as signals and images) to pass through them. That is why they are known as high-pass filters. In an image, edges are high-frequency components. The kernels we use in high-pass filters boost the intense components in an image. That is why when we apply high-pass filters to images, we get the edges in the output.

Note:

You can read more about high-pass filters at https://diffractionlimited.com/help/maximdl/High-Pass_Filtering.htm. Another type of signal filter is band-pass filters, which allow signals in a range (or band) of frequencies to pass through them. These filters allow us to highlight the edges in images and reduce the noise by using blurring at the same time. You can read more about them at https://homepages.inf.ed.ac.uk/rbf/HIPR2/freqfilt.htm.

OpenCV has a lot of library functions that implement...

Working with the Canny edge detector

The Canny edge detection algorithm was developed by John Canny. Canny's algorithm heavily uses the concept of high-pass filters. It has multiple steps.

Note:

You can read more about the Canny edge detection algorithm at http://homepages.inf.ed.ac.uk/rbf/HIPR2/canny.htm.

OpenCV has the cv2.Canny() function, which offers Canny's algorithm. The following are the steps of the algorithm:

1. A Gaussian kernel with a size of 5 x 5 pixels is applied to the input image to remove any noise.
2. Then, we compute the gradient of the intensity of the filtered image. We can use the L1 or the L2 norm for this step.
3. We then apply non-maximum suppression and identify the candidates for the possible sets of edges.
4. The final step is the operation of hysteresis. We finalize the edges depending on the thresholds passed to the images.

   Note:

   You can read more about the L1 and L2 norms and non-maximum suppression at http://www.chioka.in/differences...

Harris corner detection

OpenCV has the cv2.cornerHarris() function for detecting corners. Its arguments are as follows:

• img: The input image, which must be grayscale and have the float32 type.
• blockSize: This is the size of the neighborhood considered for corner detection.
• ksize: The aperture parameter of the Sobel derivative used.
• k: The free Harris detector parameter used in the equation.

The following is an example program that implements Harris corner detection:

import cv2
import numpy as np
import matplotlib.pyplot as plt
img = cv2.imread('/home/pi/book/dataset/4.1.05.tiff', 0)
img = np.float32(img)
dst = cv2.cornerHarris(img, 2, 3, 0.04)
ret, dst = cv2.threshold(dst, 0.01*dst.max(), 255, 0)
dst = np.uint8(dst)
plt.imshow(dst, cmap='gray')
plt.axis('off')
plt.show()

In the preceding program, we coverted the image into 32-bit float format and then we fed it to the corner detection function. Then, we threshholded the...

Exercise

To practice what you have learned in this chapter, explore the HoughLinesP(), goodFeaturesToTrack(), and FastFeatureDetector() functions in OpenCV for detecting various features. Write programs using these functions to detect lines using probabilistic Hough transforms and other features.

Summary

In this chapter, we learned the concept and demonstration of high-pass filters. We applied high-pass filters on images to obtain various results. We also demonstrated the various techniques for detecting features, such as corners, lines, edges, and circles. All of these feature-detection algorithms rely on high-pass filtering. Canny's algorithm for edge detection uses Gaussian high-pass filters. The Harris corner detection algorithm uses Sobel spatial derivatives. All of these geometric feature-detection algorithms are routinely employed in real life in industrial automation, smart vehicles, and robotics.

In the next chapter of this book, we will learn the concepts and demonstrate the restoration of degraded images; the segmentation of images; k-means clustering of one-, two-, and multi-dimensional data; image quantization using k-means clustering; and the estimation of a depth map in detail.