Saving images using lossy and lossless compression

This recipe will teach you how to save images. Sometimes you want to get feedback from your computer vision algorithm. One way to do so is to store results on a disk. The feedback could be final images, pictures with additional information such as contours, metrics, values and so on, or results of individual steps from a complicated pipeline.

Getting ready

You need to have OpenCV 3.x installed with Python API support.

How to do it...

Here are the steps for this recipe:

1. First, read the image:

img = cv2.imread('../data/Lena.png')

2. Save the image in PNG format without losing quality, then read it again to check whether all the information has been preserved during writing onto the disk:

# save image with lower compression—bigger file size but faster decoding
cv2.imwrite('../data/Lena_compressed.png', img, [cv2.IMWRITE_PNG_COMPRESSION, 0])

# check that image saved and loaded again image is the same as original one
saved_img = cv2.imread(params.out_png)
assert saved_img.all() == img.all()

3. Save the image in the JPEG format:

# save image with lower quality—smaller file size
cv2.imwrite('../data/Lena_compressed.jpg', img, [cv2.IMWRITE_JPEG_QUALITY, 0])

How it works...

To save an image, you should use the cv2.imwrite function. The file's format is determined by this function, as can be seen in the filename (JPEG, PNG, and some others are supported). There are two main options for saving images: whether to lose some information while saving, or not.

The cv2.imwrite function takes three arguments: the path of output file, the image itself, and the parameters of saving. When saving an image to PNG format, we can specify the compression level. The value of IMWRITE_PNG_COMPRESSION must be in the (0, 9) interval—the bigger the number, the smaller the file on the disk, but the slower the decoding process.

When saving to JPEG format, we can manage the compression process by setting the value of IMWRITE_JPEG_QUALITY. We can set this as any value from 0 to 100. But here, we have a situation where bigger is better. Larger values lead to higher result quality and a lower amount of JPEG artifacts.