Roxame Home - Diccan Home

Classes and regions

Classes

Main commands

classify takes the present image (img)
classify2 ...
showClasses : modifie la valeur de chaque pixel, créant une teinte uniforme pour chaque classe, et l’affiche
classe_to_region (or ctr) makes a segmentation from the existing classes
declass suppresses the existing classification

Philosophy

General considerations about classes, see diccan item.

A class is a subset of an image, containing all the points sharing a common feature. You can for instance consider the set of colors as a set of classes. But, generally, it is of little interest, and we look for more restrictive classifications. Some of the operations we have shown about colors create by themselves clasifications (notably all the palettes).

You can also look for "meaningful", "semantic" classifications. For instance classify has been built to recognize subject types : country, city, interior, head, academy.

 Note that !

- the pixels contained in a class can be placed anywhere in the image, whithout being joined.

- a pixel can belong to several classes

When Roxame classifies the pixels, it does it with the creation of integers array of the same size as the image, and noting for each point the classes where it belongs.

Tehnical hints

The code ; A_Measure

1. class computation for each pixel

In a first pass, all the posisioins of class[][] are set at 0
1. blues add 1
2. lights: add 2
4. vivid add 4
8 greens add 8
16 carnations add 16
neutrals add 32
grays add 64
if none of theses classes has been found, we are in the gray clases, from 1000 to 1003

then tabclass array, from 0 to 6

Main variables

int[][] classe. the classe of each pixel in the migage
int[][]tablass value of each class (??)
int ecl
int[]classecard, number of pixels in the class
int[][][][]tabval
floar midiangrid, medianuf
boolean mediantag$int lobpic, hobpix
int[][] semclass (on the total width of the image)

The basic array is classe[x][y].

Regions

Main commands

nbregText : displays the number of regions

Segment0 : segmentation on the displayed image. Note : to get valuavble results, this segmentation must not be applied to a complex iemage (notably a photographic picture). Use it for simple images, got direcdtly of after filters applications. Our other segmentations are in fact the combination of simplifications with a final call to SegmentO.

Segment1 : begins by a classification an aligns the pixels on their classes, then applies Segment0
Segment2 : sorts the colors on 8 classes and applies Segment 0.

Segment3 :simplifies by pixelization (1/2), VGA paletete ane median. More brutal and less subtle than Segment2.

Segment6 : segmentation après une classification conçue spécialement pour le portrait

reseg divides the number or regions by 10
deseg suppresses the segmentation
morphsegreg is a kind of filter, or image generator, but with poor results

Régions particulières :

AF1 , AF2 AF3 AF4 AF5 AF6 AF7 AF8
C1P C2P C3P C41P C42P C43P C44P
C4P
HLP HRP HWP LRP LVP LWL LWP
R0
RVP
alphonse
centerP
high_third
low_third
middle_third
random_reg
tessel4_1 tessel4_2 tessel4_3 tessel4_4
tessel9_1 tessel9_2 tessel9_3 tessel9_4 tessel9_5 tessel9_6 tessel9_7 tessel9_8 tessel9_9

Philosophy

General considerations about regions and segmentation. See diccan's segmentation item.

A region is a connected subset of pixels.
Segmentation is the partition of the image into a set of regions. We do that with the creation of an array of integers giving for each pixels the number of the region it belongs to.

Beware : segmentaion may take a lot of time (more than one minute on a 640x480 pixels).

The major use of regions is that they afford to apply operations separately to each region, notably filters, grdations and geometric forms generation.

A simplfied image of the regions, with arbitrary colors, is displayed in the low part of the Roxame's scree

Regions are numbered from 1. At start, of after a desegmentation, there is only the 2 oregion. The selection of region is done by the function R, which notably computes the coordinates of the rectangle circumscribing the region. A call to region 0 points on the whole of the region.

At the the limit, we could have as many regions as pixels. Good segmentations, in our experience, have typically some thousands of regions.

Technical hints

The code ; Roxame

Main variables

int[][]region
int regone, nnbreg
int nbregutil : number of regions operational at a given time
boolean[] regocc
int cardreg,
int xd,yd, xf, yf : coordinates of the rectangles circumscribing the region