Tire tread classification problem is primarily of use in two areas. First and foremost, forensics. It could come of great use in a case involving motor vehicles to be able to at least determine what kind of tire has been in a crime scene, and subsequent analysis of the possible vehicles wearing that same tire. Secondly, it could also be used by tire manufacturers to prevent copyright infringement of a tire tread pattern. We want to think that given a proper use, this technique could be of help to forensic labs around the globe, and perhaps, on some extent, it could also be used on other fields of image recognition.Real world applications are always the highest motivation you can get. Being my first research paper I felt that focusing on a concrete and concise goal was of real help. We looked for a way to provide the user to identify a tire model, given a partial tire tread image from the ground. At least we need a way to guide him along and choose a few of the extensive database with probable solutions to the tire tread he is looking for and not having them to look for in a three hundred or more tire tread images. The concept of a partial tire tread image, is the key to this work, and what it makes it unique amongst other systems. Most existing systems work with a complete tire tread, but in most cases, all you can get is a small part of the complete tread. This is why we focused on proposing a solution able to work with just a part of it. On the other hand this also was also one of the major pitfalls we encountered. Not having the complete information of what we are looking for made all the existing research on tire tread matching unusable, leading us to start from the ground-up. We are looking for tire tread recognition in its full extent. We believe that any step we take to our final goal is welcome. Given that the user should be supervising the software already he could also be the one deciding which option is the best, given a current result presented in a correct fashion for him to do so. An exact matching would definitely be great but, as in most cases on computer vision field, this is nearly impossible.