Overview / Usage

Initially user draws a rectangle around the extraction region for clothing and human hair from input image. (foreground region shoule be completely inside the rectangle). Then algorithm segments and masks the desired clothing or hairstyle iteratively to obtain an optimized result.

3D Picha Systems

1.Detects and registers the image of a person's clothing and hairstyle into a database
2.Recognizes known clothing and hair styles in a video stream using webcam
3.Mask Clothing and or Hairstlye if the clothing matches an image in the database

Methodology / Approach

Power Paring OpenVINO Inference Engine and OpenCV

The most promising computer vision background removal technique for this project is GrabCut. GrabCut is an iterative segmentation method that requires the user to specify a bounding box around the region of interest.We shall load a Gaussian mixture model, typically of 5 components to Intel's Model Optimizer intermediate representation and load to opencv.

Parameters
[in] xml XML configuration file with network's topology.
[in] bin Binary file with trained weights.
Returns
Net object. Networks imported from Intel's Model Optimizer are launched in Intel's Inference Engine backend.

We then use it to estimate the color distribution of the target and background in an iterative manner until convergence.

One property that makes GrabCut an excellent choice is the ability to provide cleaner segmented edges, owing to its border-matting feature. Given that our aim was to reduce as much irrelevant content as possible from the query image, this feature was crucial in this use case. It’s also worth noting that the GrabCut method allows human interaction to ‘correct’ the segmented results, manually indicating misclassified regions and then rerunning the optimization.

Third Party Subsystems

The 3D Picha system consists of two main subsystems:

cvservice:
cvservice is a C++ application that uses OpenVINO™.

webservice:
webservice uses the MQTT broker to interact with cvservice.

In the UI, there are three tabs:

1.live streaming video
2.Clothing and Hairstyles as input from user or from existing samples
3.analytics of fashion style history.

Technologies Used

Hardware requirements
5th Generation Intel® Core™ processor or newer or Intel® Xeon® v4, or Intel® Xeon® v5 Processors with Intel® Graphics Technology
USB Webcam [Logitech* C922x Pro Stream]

Software requirements
Ubuntu* 16.04
OpenVINO™