Image analysis and computer vision has always been important in industrial applications. With the popularization of cell phones with powerful cameras and Internet connections, they are also increasingly being generated by the users. Therefore, there are opportunities to make use of this to provide a better user experience.
In this chapter, we will look at how to apply techniques you have learned in the rest of the book to this specific type of data. In particular, we will learn how to use the mahotas computer vision package to preprocess images using traditional image-processing functions. These can be used for preprocessing, noise removal, cleanup, contrast stretching, and many other simple tasks.
We will also look at how to extract features from images. These can be used as input to the same classification methods we have learned about in other chapters. We will apply these techniques to publicly available datasets of photographs.
From the point of view of the computer, an image is a large rectangular array of pixel values. We wish to either process this image to generate a new or better image (perhaps with less noise, or with a different look). This is typically the area of image processing. We may also want to go from this array to a decision that is relevant to our application, which is better known as computer vision. Not everybody agrees with this distinction of the two fields, but its description is almost exactly how the terms are typically used.
The first step will be to load the image from the disk, where it is typically stored in an image-specific format such as PNG or JPEG, the former being a lossless compression format and the latter a lossy compression one that is optimized for subjective appreciation of photographs. Then, we may wish to perform preprocessing on the images (for example, normalizing them for illumination variations).
We will have a classification problem as a driver for this chapter. We want to be able to learn a support vector machine (or other) classifier that can learn from images. Therefore, we will use an intermediate representation for extracting numeric features from the images before applying machine learning.
Finally, at the end of the chapter, we will learn about using local features. These are relatively new methods (SIFT (Scale-invariant feature transform), the first element in this new family, was introduced in 1999) and achieve very good results in many tasks.