The idea behind this paper is to try to classify objects based on more than one differentiating feature. The ones they used where colour, shape and texture. Each feature follows a different process but all end up being vectors. At the end of the process the vector is combined (they normalize the combination) and nearest neighbour is used for classification. In the paper they made a comparison of the classification behaviour of the individual features and of the combined features.
The way they evaluated the behaviour of their resulting algorithm was a bit strange. The were creating a program the returned a list of images that were of the same type as the test image. Their results are mainly based on the 5 first images of this list. So they are saying that the algorithm was able to put the right flower within the first 5 elements of the resulting list with a certain accuracy.
Colour feature calculation.
They made emphases on the use of segmented images. They argue that the colour of the background will negatively affect the feature extraction if the flower is not segmented out of its background. They used HSV for the colour space. The used k-means clustering to create a vector with the calculated means. The search for the optimal number of clusters and found that 500 was the best value.
Shape feature calculation
The used a rotationally invariant descriptor. They use a combination of SIFT and HOG. The compute SIFT descriptors on a regular HOG-like grid. They search for the optimal cell size, step size. They obtain the vector through vector quantization. I didn’t really understand how they calculated the shape feature. (have to read more into SIFT to see if I can figure it out). No comment is made on segmentation (I’m not sure if they used segmented or non-segmented images). The idea here is to capture relations between the sub-sections of the flower: like overlapping petals and petal pointedness.
The idea behind the texture calculation is that some flowers have certain texture in their petals. Here they used an MR8 filter. The resulting calculation is rotationally invariant because they choose the maximum response over all orientations. They create the resulting vector by clustering the descriptors and frequency histograms.
This paper contains lots of helpful directions on where to look for possible solutions for our problem. One of the things that I am most positive about is the use of HSV colour spectrum in the detection. (I think the Salix Arctica females might have a very distinctive HSV behavior).