Login Register

Auto-login on future visits

Forgot your password?

perClass imaging framework

Seamlessly integrates imaging tools with statistical pattern recognition. We designed it to deliver our projects in the fields of defect detection, texture classification or object recognition.

It may significantly speed up also development of classifiers in your imaging applications!

Label painting and annotation

Easily label areas of interest in your images. Define classes by painting or clustering. Quickly process clustering results by renaming uninteresting clusters and assign meaningful names to clusters you can interpret.

See for yourself how quickly you can prepare and accurately label your training data set from a directory of images.

interactive image labeling in Matlab
image and data set connection

Image / data duality

perClass enables you to move freely between image domain and related feature space. Any image is a data set. Any derived subset can be still viewed as an image.

By connecting the image and scatter plots, you may quickly understand your data. What is that strange cluster over here - oh, it collects the edge pixels! Where does my classifier make errors? Can I visualize it in the image?

Answer such questions in few clicks, not in an afternoon.

Sparse images

In many imaging applications, you're only interested in objects in the scene, not in the background. perClass allows you to work with arbitrary pixel sets. Any subset of pixels, such as detected object, can be still viewed in the context of the original image.

You may still compute local image features, perform filtering or visualize your classifier decisions on the pixels subsets. It saves space and, even more importantly, it saves your time.

See short video

working with partial sparse images in Matlab
training classifier on multiple images in one data set

Learn from multiple images

One image is not enough to develop a robust defect classifier or tissue characterization. Robust models need to be trained from multiple images. perClass helps you to collect data from many images in one data set and train reliable and accurate classifiers.

See short video

Local image features

You can only build accurate classifiers if you start from a good quality feature set. perClass helps you to quickly characterize local image information by a broad range of extractors implemented in a high-performance C code (*).

Train texture classifiers at different scales, discriminate regions based on edge saliency or local orientation. Apply filter banks. All built-in extractors support masking, user-defined block sizes and image grids to further speed up processing of your massive images.

(*) Or use the same framework to apply your custom Matlab code to extract the most relevant features for your application.

See short video

local image feature extraction in Matlab
segment objects with connected component analysis in Matlab

Object segmentation and representation

Easily segment connected components on any classification or clustering result. Discard small regions with one click. Describe connected regions by object-level features and train object classifiers.