Graphical passwords are an alternative to usual textual passwords. They use an image as main support and image handling such as pointing position in the picture as entry mode. They can be convenient on tactile screens, more difficult for robots to mimic human behavior, and claimed to offer better memory resilience.
Since early 1990s, the literature has been rather extensive in the field. Technicolor published several papers in the field (search for Maetz and Eluard). But we rarely see a product that implements such a solution.
UK-based company, PixelPin offers such a solution. It is based on Bonder’s seminal patent (5559961). When registering, you select one image as a support and four points in the image in a given order. When answering the challenge, you have to select the four points in the initial order. To limit risks of shoulder surfing, the precision of positioning is rather fine (at least on a computer). After 5 attempts, the account is locked for 15 minutes. Reset sends a reset token via the email used to register.
To increase memory resilience, and to ease the positioning you should select a picture with clear identified salient points else you will be quickly locked out. Of course, using too obvious salient points reduces the space of “keys” to explore.
The main issue is the network effect needed for such solution. It will be efficient if the sites are common and often visited, else your memory will fade. Unfortunately, I did not find many sites using PixelPin. The startup was launched beginning last year.