EyeRecognize's NuditySearch - Recognizing nudity is a highly complex problem. NuditySearch tackles this problem by recognizing anatomical attributes and determining if there is nudity or suggestive imagery in images. The service is useful for website or app owners looking to provide site safety, ad safety, and online profile verification. We provide 3 different settings of varying sensitivities depending on your tolerance for exposed skin (i.e. - lingerie and bikini).
This API accepts the URL to an image and returns the following:
"suspect": True/False depending on if engine detects nudity.
"monochrome": true if the image is black /white or has modified by color filters
"highLight": true if the image has be over exposed
"qualityLevel": value of the jpeg compression quantization level, higher levels of compression can affect the quality of an image
"noiseScore": value from 0 - 100 representing the amount of random noise, likely affecting the quality of the image
"score": value from 0 - 100 representing the engines confidence in the result. The score can be used in lieu of or in addition to the suspect field.
"imageDimensions": pixel width and height of the image.
We use the combination of shape and body part segments with a multi-layer classifier to provide for recognition of genitals, breasts, and buttocks, as well as a general classification of nudity when sexually suggestive body regions are displayed.
On the Loose and Standard sensitivity setting, the engine now takes into account the environment and surroundings of the subject matter. Because of the fine line between lingerie and bathing suits, intent can become an important perspective. By taking into account the surrounding environment, our classification process now includes more situational awareness to reach a conclusion.
Sensitivity parameter now called Setting
ImageVision allows customers to define their tolerance for nudity vs false positives by adjusting the solution sensitivity settings to fit your application of the technology.
There is an explicit trade off related to the definition of nudity and the ability for any computer technology to recognize it. When a strict definition is applied, the technology is more likely to identify borderline cases as nudity or non-nude images as nudity making a false positive identification. With a more relaxed setting, images may be misidentified as “clean” when in fact; there is nude imagery or a false negative identification.
The different sensitivity settings allow for different degrees of false negative and false positive tolerances of the solution allowing for the definition of nudity to fit the application of the technology.
An example of this in action is the following image - http://i.imgur.com/Ry2WlxD.jpg
At Settings 1, the image is marked CLEAN. At Setting 2 and 3, the image is marked NUDITY.
Please let us know how the solution is working for you or how we can improve.
The EyeRecognize Team