Update - 250607

- Completed block segmentation to separate individual blocks from the slides, notably to extract visual elements such as graphs, charts and tables from the slides (see Fig1).

- Using the original image to segment instead of the extracted slides, the problems of blurry and non-complete images can be solved (see Fig 1)

- Found a program pdf2Text to extract all the texts from a pdf slide set into text files. Using that, the next step is to think about what are the key words to be linked to a particular visual element.

- Started on reading up of how google carries out queries and the data structures used to store the index.

TO DO:
- Which key words to be linked to a visual element in a slide?
- To identify what kind of images the segmented images are: text, tables, charts, pictures etc.
Fig. 1: The segmented visual elements (example) cropped from the original slide to prevent any loss of quality.

Update - 15/6/07

Updates

- Successfully separated the background image from a series of slides (see Fig 1)
- Extracting of foreground images from slides (see Fig 2)
- Detected some memory leaks which caused program to be slow

- Background extracted may contain missing or distorted pixels which may cause some foreground images to be unreadable.
- Tried to smooth the images extracted by:
a) Taking the dominant colour of the surrounding pixels
b) Taking the average rgb values of the surrounding pixels
- Results unsatisfactory

- Started on block segmentation
- Successfully implemented a code to push all the pixels to the left (see Fig. 3) and to the bottom to form necessary histograms for segmentation

TO DO:
- Refine the background / foreground extraction method to produce better quality output
- Analyse the histograms produced in order to segment the images





Figure 1










Figure 2












Figure 3