Martin McBride - Image Processing in Python

This part of the book covers bitmap imaging.

This chapter will cover the basics of what a bitmap image is. Later chapters will cover:

• How computers represent colour.
• Colour models.
• Colour resolution.
• How image data is stored in memory.
• Transparency.
• Image compression.
• Image file formats.
• Colour management.

You most likely already know what a bitmap image is. Almost any image you see on the web will be a bitmap image, and you have probably used your smartphone or digital camera to capture photographs as bitmap images.

You might be more familiar with alternative names - raster image, or pixel image. They mean the same thing as bitmap image. They are sometimes also called JPEG images or PNG images, named after specific image file formats.

You probably also know that a bitmap is made up of pixels - they can be thought of as tiny coloured squares that make up the image. They are normally too small to see but become visible if you zoom in too far and the image becomes pixelated.

This chapter presents an overview of the characteristics of bitmap images, in preparation for the remaining part of this section that looks at bitmap images in detail.

A real-world scene has an almost infinite amount of detail. If you look out at, for example, a boat by a lakeside, it goes beyond the detail your eye can see. You could walk up to the boat and look at it in virtually unlimited detail.

A bitmap image has a finite amount of detail. If you took a digital photograph of the boat, the camera would convert it into an array of pixels. Each pixel represents a small part of the image.

If the pixels are very small together, we cant distinguish the individual, and the image looks similar to the actual scene. If they are larger, we see the image as a set of pixels rather than a natural image, as this illustration shows:

In practical terms, at a viewing distance of 40 cm, the eye can resolve objects that are about 0.1 mm apart. So for example, imagine a sheet of paper with two thin lines drawn 0.1 mm apart. If you held the paper 40 cm in from of your face, assuming you have normal eyesight, you might just about be able to see that there were two separate lines. Any further away and they would just look like a single line.

This means that if you wanted to print an image on a page so that the eye couldnt see the individual pixels, you should aim for a pixel resolution of 10 pixels per mm (about 250 pixels per inch) or better. So for a printed photograph or 15 cm by 10 cm, you would want an image of at least 1500 by 1000 pixels. That would be a 1.5 megapixel (MP) image.

In reality, you would probably want a higher resolution than that:

• To allow you to print larger photographs.
• To allow you to crop a photograph to show just the subject.

Most modern digital cameras support image sizes of 6 MP or higher.

However, very large pixel sizes are not always as useful as they might seem:

• For printing very large images, such as posters, they are normally viewed from further back than 40 cm, so there is no need to have 0.1 mm spatial resolution.
• At very high resolutions, factors such as camera shake will blur the image, so there is little point in taking a very high-resolution photograph without a very solid tripod.

Each pixel in a bitmap image has a specific colour. There are various ways we might represent a colour in an image.

The most common way to represent a colour is as three separate components, the amount of red, green, and blue light that make up the colour. As we will see, this is based on the way the human eyes perceives colour. However, we sometimes use alternate methods, including:

• CMYK - used to represent colours for printing.
• HSL - used in art and design as an intuitive way to select related colours.
• Perceptual colour spaces such as CieLAB used to create very accurate colours.

An important consideration is how precisely we need to represent each colour in an image. As rough a rule of thumb, we can detect variations in colour of about 1%. Most modern systems use 8 bits per channel (for example, three integers between 0 and 255 to store the red, green, and blue values). This gives a precision of better than 0.5%, which is adequate for most uses.