Image Processing: Why Lena doesn’t matter anymore.

In image processing, a worn out image, is one that does not truly reflect the real-world algorithms for which it is used for testing purposes. There are a series of these clichéd images: mandrill, peppers, cameraman. But the most notorious of these is certainly “Lena”. Lena is the name given to a portion of a centerfold picture of Lena Sderberg, a Swedish model posing in the November 1972 issue of Playboy magazine. Below are two images of Lena. The first is the standard image which has circulated since it was digitized in 1973 (USC-SIPI Image Database), the second a grayscale version of the image more commonly used in illustrating image processing algorithms.


The January 1996 IEEE Transactions on Image Processing has a note from the Editor-in-chief who says that “image contains a nice mixture of detail, flat regions, shading, and texture that do a good job of testing various image processing algorithms“. This might be true, however with the vast diffusion of digital cameras and the widespread use of the internet, producing a colossal inventory of digital images of a variety of subjects, it is hard to believe that Lena is the solitary example available to researchers.

Over forty years since the picture was digitized it can still be found in literature on image processing used to test algorithms. Consider the case of colour image segmentation. Why would a person want to segment Lena? It is possible they might want to design algorithms to segment features from a colour video input stream to guide a robot. Is the robot going to run into a picture of Lena anytime soon? It doesn’t even make sense in the context of a facial recognition application. Lena has been used in the context of testing bi-modal thresholding algorithms. Here’s the histogram of the grayscale image:


Hardly bimodal. In fact, the  histogram of intensity distributions is multi-varied, there is no clear division of intensities. So why use this image to test a bimodal thresholding algorithm? It’s complete nonsense. Go to Google Scholar, search for “Lena” and “image processing” – what you will find is that even in 2017, people are publishing papers using Lena as a test image in their experimental sections. It’s almost beyond ridiculous.

Look, Lena makes a nice historical image – an image that like its fellow images in the USC-SIPI database helped the fledgling field of image processing evolve. But now, it’s time to archive Lena and look to testing algorithms with more realistic images.
Want to learn more about worn out images? Read this paper I wrote a few years back: “Worn-out images in testing image processing algorithmsPDF.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s