Monday, August 13, 2012

Initial interface

Hello, this blog post describes the current and the envisioned functionality of the interface to process specimen records.

DEMO: link

Short tutorials:

  1. To start, click on "New Annotations" button on the top right
  2. Draw a bounding box around a text region
  3. A new window will pop up
    1. Here you can select whether this text is machine-printed or handwritten
    2. In the text area, input the text within the bounding box
    3. Press "Done" when finished.

Figure 1. Click on the "New annotations" to start the process
Figure 2. Drawn bounding box and information filled. Click "Done" to finish.

Figure 3. A complete example. The text area in the right illustrates the information that is going to be sent back to the back-end server of choice.
Interactions:
  • Draw bounding boxes around text
  • Resize and move the bounding boxes
  • Input and modify the texts for the drawn bounding box by clicking on the bounding box
  • Delete bounding box

Current benefits:
  • Allows easy annotations of texts and data
  • Easy interface with other system (Mechanical Turk, ZooUniverse, etc.)
  • Easy access - web interface - no installation required.

Before reading the next part, please take a few minutes and experience the interface. The next section will make more sense.

Upcoming features:
  • Using computer vision and machine learning to streamline the process
    • Use automatic text detection for automatic bounding boxes localization
    • Use OCR to automatically fill in the text box section
    • Use word spotting algorithm to deal with tough cases
  • A time measurement mechanism to evaluate the performance of the interface.
In the next blog post, we will describe the use of text detection in assist of automatically localizing bounding boxes.