Computer vision
Budget: $30 – $250 USD
1.Implement the global colour histogram from using a Euclidean distance metric.
Experiment with different levels of RGB quantization
2.Compute precision recall (PR) statistics for each of your experiments, e.g. PR for the top 10 results.
Plot the PR curve. If you are defining similarity in terms of object categories, compute a confusion
matrix. Discuss and analyse your results e.g. which experiments were most successful (which images
worked well; why do you think this is given your descriptor choice).
3.Implement a basic BoVW system using a sparse feature detector (e.g. Harris or the SIFT keypoint
detector) and a descriptor (e.g. SIFT descriptor). Use k-Means to create the codebook. Compare the
performance with other descriptors you have tried.
4.Apply an SVM to try to decide the category of the image e.g. “bike” or “sheep” based on the
descriptors you have extracted from each image. (This is classification)
You are provided with basic code for iterating through files in the Microsoft dataset
a structure code is already given
Experiment with different levels of RGB quantization
2.Compute precision recall (PR) statistics for each of your experiments, e.g. PR for the top 10 results.
Plot the PR curve. If you are defining similarity in terms of object categories, compute a confusion
matrix. Discuss and analyse your results e.g. which experiments were most successful (which images
worked well; why do you think this is given your descriptor choice).
3.Implement a basic BoVW system using a sparse feature detector (e.g. Harris or the SIFT keypoint
detector) and a descriptor (e.g. SIFT descriptor). Use k-Means to create the codebook. Compare the
performance with other descriptors you have tried.
4.Apply an SVM to try to decide the category of the image e.g. “bike” or “sheep” based on the
descriptors you have extracted from each image. (This is classification)
You are provided with basic code for iterating through files in the Microsoft dataset
a structure code is already given