Options

# Unflatten image data, process it one image at a time.

I have a data set that is a collection of labeled images I want to process.

For the sake of example, let's say that each image is 32 rows and 32 columns of intensity values (intensity is an integer between 0 and 255). 32 x 32 = 1024, so each image has 1024 pixels and each pixel is a single integer.

Each example in the data set set represents a single image and its label:

column 1: nominal label of the image

columns 2-1025: Intensity values of each row are appended to one another to create a single, long record

I want to perform PCA on each image, row-by-row, and keep the 5 most significant principal components of each row. Then I want to string these principle components together to create a feature vector for the image.

The transformed data set would look like:

column 1: label

columns 2-161: real numbers (5 real numbers for each row of an image, and an image has 32 rows).

I'm not sure how to do this in RapidMiner. Can anyone point me to some examples or describe a solution?

For the sake of example, let's say that each image is 32 rows and 32 columns of intensity values (intensity is an integer between 0 and 255). 32 x 32 = 1024, so each image has 1024 pixels and each pixel is a single integer.

Each example in the data set set represents a single image and its label:

column 1: nominal label of the image

columns 2-1025: Intensity values of each row are appended to one another to create a single, long record

I want to perform PCA on each image, row-by-row, and keep the 5 most significant principal components of each row. Then I want to string these principle components together to create a feature vector for the image.

The transformed data set would look like:

column 1: label

columns 2-161: real numbers (5 real numbers for each row of an image, and an image has 32 rows).

I'm not sure how to do this in RapidMiner. Can anyone point me to some examples or describe a solution?

0