how to match bra just by pic of the chest using image processing??

hi..my name is matan. im doing a project and i need to match a bra (size only) for a woman. The woman need to take a pic of her chest and then using image processing i want to run gui code that give me back her bra size... thx!

Answer 1

[Walter Roberson]

It is not something that can be calculated with a single picture. You need a minimum of two different perspectives, and both of them need to have scales (lines of known length) in them.

You have some distinct issues:

1. identifying the breast volume;
2. identifying the band size (circumference of the torso at the appopriate location); and
3. converting the above two measurements into bra size

There are competing hypotheses about how to best measure breast volume; I have included a link to a couple of research articles. The first did comparitive research; the second I do not think was researched well enough to check it against the alternatives. The third one, the Breast-V system from Italy, is the one that is the most accurate that I know of, but it also requires the most measurements.

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3076010/

http://www.ijps.org/article.asp?issn=0970-0358;year=2006;volume=39;issue=1;spage=13;epage=16;aulast=El-Oteify

http://www.diepflap.it/en/11.7/breast-volumeweight-calculator.htm

The next thing you need to know is that cup size is basically a measurement of breast volume. Cup size is not a matter of breast projection. The assumption is that different breasts of the same volume will squish around to conform to the boundaries of the cup. There are some kinds of breasts for which that is true, but there are others where it is unjustifiable. The firmer or denser or more fiberous the breast, the less it is likely to easily squish. But those are the same conditions under which it is easiest for you to estimate volume by dual perspectives: the less-dense or less firm breasts that deal well with that system are probably going to have "flowed" in pictures and be much more difficult to get a simple volume measurement of.

There is a relationship between band size and cup size that holds fairly well from band size 30 (US) to 38 (US) and in the corresponding European sizes, but increasingly fails after that. The relationship is that as you move up to the next larger band size (2 inches in US/Canada, 5 cm in European) that the any particular volume is in the next lower cup size. For example, if you needed cup (breast) volume labeled as C for US band size 34, then the same volume would be labeled as B for US band size 36. Unfortunately by band size 42 (US) the progression gets more and more inaccurate and any given larger person, in changing bandsize by one increment for the same bra model, could end up needing to use the same cup size or could end up needing to go down two cup sizes.

Formally speaking, though, what cup size really measures is the size of the circular opening for the breast. A manufacturer could give more or less volume in the cup, or more or less stretchibility for the cup, and still label it with the same cup size, and this can vary from bra model to bra model, so you cannot really trust cup size, but treating cup size as talking about volume "works" most of the time. But because cup size is formally based upon size of the opening, that corresponds to width of breast base, which is something you might have hopes of measuring by photograph. Unfortunately, you start running into perspective problems especially on larger people. There is a fraction of people (mostly petite XX chromosome people) whose rib cage curves inward enough that the breasts are pointing mostly forward, but for everyone else of Tanner Stage II or beyond, the centroid of the base far enough away from the center of the chest that the breasts angle outward toward the left and right in a "V" and the base typically needs to be measured along the curve of muscle or else from an angled view that is neither side view nor front view. The larger the ribcage or the larger the volume of breast, the less likely it is that you would be able to accurately estimate the breast base width by front or side photograph.

If you search online, you will find all kinds of stores and bra resources that tell you how to calculate bra size based upon either two or three measurements, typically a mid-bust circumference and an underbust circumference. It turns out that those two measurements together are not very accurate at pinning down bra size, with it being not unusual for people to need to move up or down one band size and up or down one cup size. If you see a measurement procedure that talks about subtracting anywhere between 1 inch and 5 inches, then it turns out that procedure is basically only aimed at women from band size 30 (US) to 36 (US) who have noticable waists, and it is almost always wrong for women who are 38 (US) or larger or who have broad rib-cage. If you see a "subtract 4" then run away: that was really only ever valid for Warner bras produced in the USA.

The US and Canada (and even UK) band measurement systems are more or less crap. The European band measurement system is much more consistent. With one exception: Triumph band sizes are almost always significantly lower than what they are labeled with even when you use their European measurements. (I have had a small number of people tell me that Triumph is the only manufacturer whose bandsize they can trust, but most of the opinions I have read say instead that Triumph is almost always the wrong band size.

Sorry I am falling asleep as I type this.

[Image Analyst]

Who knew Walter was such an expert on bras? I bet this post gets the record for number of views from those hoping your posted photos, as people who ask for image processing advice usually do. What are your plans for this program? Run a large university study and then turn it into a web service so women can buy bras online? Good luck with that.

[Star Strider]

You probably need a laser scan of the woman’s torso to do this correctly.

[Walter Roberson]

Another relevant paper on volume calculation:

http://journals.lww.com/plasreconsurg/Abstract/2006/08000/Breast_Volume_Determination_in_Breast_Hypertrophy_.5.aspx

Although the R value for it is 0.89, the 11 percent of variation that is left is enough to make a difference of +/- 1 cup size for even moderate breast volumes. The standard used by plastic surgeons for implant / augmentation purposes is that 1 cup size corresponds to 100 ml (100 cubic centimeters), which is why implant and reconstructive device manufacturers typically sell in either 100 ml or 50 ml increments.

See also http://journals.lww.com/plasreconsurg/Citation/1998/09020/A_New_Method_for_Determining_Bra_Size_and.56.aspx

Another item to note is that appropriate bra size is strongly influence by the amount of squishable torso fat. Bras that need to provide support get only roughly 10% of the support from the shoulder straps, and get the rest of the support from the band being firm around the torso. But for those with more torso fat, the required firmness is probably going to push in and displace some of the fat, and there is no good answer for what the "right" firmness is, other than the guideline that it should not hurt and it should not be tiring. Too little firmness and the band might not be providing enough support for larger volumes; too much firmness and it hurts and tires and displaces the fat in displeasing ways. The "muffin top" look can be a dreaded problem for some people. But it is also a real problem in choosing cup size, because the place such fat is most significantly displaced is right along the outside edge under the armpits, and it typically gets displaced toward the front, moving the fat into the place that needs to be covered by the cup base, requiring a bigger opening (larger cup size) even though the fat might not technically speaking be breast volume.

Star Strider mentioned laser scanners. I am aware of at least three competing manufacturers of scanners for use in plastic surgery clinics, with two of those being strong competitors and the third being somewhat behind in sales when I was researching this a few years ago. Visible light laser is good for the systems that are offered for visualization purposes (the client gets scanned and then the software is used to show the client what they would look like with different augmentation sizes), but if you want a tissue analysis so as to be able to distinguish breast tissue from other tissue, or to help make predictions about tissue mobility under band tension, then you are better off going for ultrasonic or infrared scans.

With regards to photographs: "selfie" photographs typically require that the person be holding up the camera, and that significantly distorts the distribution of tissue and makes it much harder to get an accurate assessment of breast base. The photos should really be with the person with their arms down resting lightly by their sides.

My expectation of a system based upon single user supplied photographs is that it would at best be useful for young slim xx chromosome people who have no intersex conditions, no visible chromosome mosaicing, no significant chimera effects, no Disorders of Sexual Differentiation (DoSD). That is the high-volume low-profit "fashion bra" market (where, frankly, a good fit is considered a bonus and not really expected by consumers), together with the high-volume ultra-low-profit "basic bra" market where $5 bras from China have significant market share. One you get out of those markets, you have major competition based on brand name (in particular, Victoria's Secret) for the average consumer. But the larger per-unit marginal profit is in getting the right fit for those with broader rib cages or more torso fat or larger breast volumes. A lot of people knowingly "make do" with the wrong bra size "since nothing ever fits right anyhow, I might as well buy cheap", but some of the harder-to-fit find their way to specialty shops and services and end up paying hundreds of dollars for even basic bras. There are a number of different online communities where people get really serious about searching out the less common sizes and comparing them style-by-style. See for example bratabase (caution: site includes plenty of pictures of bras being worn)

Now of course you could do one week undergrad project without much research and with limited accuracy, knowing that the code will be thrown away as soon as it is marked, but if anyone reading this happens to be thinking about making real system intended for real use, then I advise doing real research, reading the science papers, hunting around in Google Scholar, especially (but not exclusively) in the field of Plastic and Reconstructive surgery, read sites such as realself (caution: site includes plenty of pre- and post-op plastic surgery pictures). You would need that material to properly understand how breasts are shaped and positioned and scientifically accurate volume measurements. And if you do take on such a project seriously, remember to take into account intersex, DoSD, trans conditions, and gynecomastia (breast development on males; autopsy studies suggest up to a 60% prevalence for some grade of it) (caution: any serious article about that will include pictures.)

What I would not even consider doing is a system based upon "candid snapshots" of a clothed person. I have seen in the past a couple of web sites that purport to estimate bra size from standard clothed pictures like that. The accuracy on those sites was always astonishingly poor.

Answer 2

[larry frederman]

hello matan..

Well you'll have to write your own version of either imfindcircles() or bwlabel(), and regionprops(). You might need to threshold your image to binarize it. Then, you can find connected components algorithms on the web to do the labeling of your binary image. To get the diameter, sum of the number of pixels in each blob and use the formula for area. Diameter = sqrt(area*4/pi). after that calculate all the prmiters (daimeter, premiter) and then chek on the web the sizes of bra and do some kind of Comparison between the Measurements and the bra data.

hope i helped you matan.

for more help contact me.

larry

[matan marciano]

thank you very much!

I will try to do so.

[Image Analyst]

I don't see how this could possibly work.

[Walter Roberson]

The strategy described here would, at best, get you the area covered by the bra from the perspective of the picture. The area covered by the bra has a weak link to bra size, at least until you get beyond roughly 800 cc per breast (38D for example), in which case what starts to increase more rapidly is band width.

[Image Analyst]

I imagine imfindcircles() could find the nipples alone, but not the full breast. The whole breast is not outlined in a circle that is visible. so determining the area of the full breast is not possible. And the suggestion of thresholding is also doomed to failure I predict. It's not easy to determine exactly where the "top" of the breast is, and in any case, it's probably a judgement call. I don't know what circles larry thinks could be found that predict bra size, or what he thinks could be thresholded.

I also doubt whether any woman would take a naked picture of her chest and send it off somewhere (a web site) for image analysis which would then determine the "optimal" bra size, which could then be purchased (but I could be wrong).