For our shape classification work [1],[2],[3] along with standard shape datasets, we experimented with some datasets which we created specifically for this classification task.

A. Fighter plane shapes:

The fighter airplane shape database included Mirage, Eurofighter, F-14, Harrier, F-22 and F-15. Since F-14 has two possible shapes, one when its wings are closed and another when its wings are opened, total number of shape classes are seven. Each class includes 30 shape samples. Shape database was created by taking digital pictures of die-cast replica models of these airplanes from top. Pictures were captured at 640X480 resolution, and were segmented. Contours of the segmented planes were used for training and testing of the classifier.

Airplane shape classes: (a) Mirage, (b) Eurofighter, (c) F-14 wings closed, (d) F-14 wings opened, (e) Harrier, (f) F-22, (g) F-15.

B. Vehicle shapes:

Approach discussed in our ICIP 2005 paper [4] was implemented and applied to outdoor videos to extract these shapes. The vehicles were classified into one of the four classes: sedan, pickup, minivan, or SUV. Videos were captured at resolution of 320X240. As object extraction approach used does not deal with shadows, the extracted car shapes are distorted in the bottom half due to shadow. For each class, 30 samples were extracted from the video. For this dataset shape samples show larger within-class variation, as shapes of vehicles of different makes and models vary. Also, the contours extracted show higher degree of deformation due to the shadow problem in object extraction.

Vehicle shape classes: (a) Sedan, (b) Pickup, (c) Minivan, (d) SUV.

C. Subset of MPEG-7 CE Shape-1 Part-B

MPEG-7 CE Shape-1 Part-B data set includes 1400 shape samples, 20 for each class [5]. Part of this data set was used in HMM based shape classification experiments in [6], [7]. We use the same shape classes, however we carried out two different sub-experiments: one with all the shape samples and the other similar to [6], [7] where only 12 samples from each class are included . We choose following seven classes: Bone, Heart, Glass, Fountain, Key, Fork, and Hammer. The shape classes are very distinct, but the data set shows substantial within-class variations.

Part of MPEG-7 CE Shape-1 Part-B shape classes.

Download

The datasets can be downloaded here. The archive contains 4 different folders plane_data (Fighter plane shapes), car_data (Vehicle shapes), bicego_data (Subset of MPEG-7 CE Shape-1 Part-B used by [6], [7]) and mpeg_data (Subset of MPEG-7 CE Shape-1 Part-B used by us). In each folder, Cartesian coordinates of each point on the perimeter of the shape are stored as MATLAB 6.5 data files named as Class%d_Sample%d.mat . Each file can be loaded into MATLAB workspace with simple load command. This will create a variable x in MATLAB workspace of size N X 2, where is the perimeter of the shape. These shapes can be easily visualized by executing  plot(x(:,1),x(:,2)). If you utilize our datasets then please refer our TIP paper[1] in your work.

References:

[1] N. Thakoor, J. Gao, S. Jung, "Hidden Markov Model-Based Weighted Likelihood Discriminant for 2-D Shape Classification",  IEEE Transactions on Image Processing, Volume 16, Issue 11, Nov. 2007, Page(s): 2707 - 2719.

[2] N. Thakoor, S. Jung, and J. Gao, “Hidden Markov model based weighted likelihood discriminant for minimum error shape classification,” in Proc. IEEE Int. Conf. Multimedia and Expo, 2005.

[3] N. Thakoor and J. Gao, “Shape classifier based on generalized probabilistic descent method with hidden Markov descriptor,” in Proc. IEEE Int. Conf. Computer Vision, vol. 1, 2005.

[4] N. Thakoor and J. Gao, “Automatic video object shape extraction and its classification with camera in motion,” in Proc. IEEE Int. Conf. Image Processing, vol. 3, 2005, pp. III–437–40.

[5] Shape data for the MPEG-7 core experiment CE-Shape-1. http://www.cis.temple.edu/~latecki/TestData/mpeg7shapeB.tar.gz

[6] M. Bicego and V. Murino, “Investigating hidden Markov models’ capabilities in 2D shape classification,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 26, no. 2, pp. 281–286, Feb 2004.

[7] M. Bicego, V. Murino, and M. Figueiredo, “Similarity-based classification of sequences using hidden Markov models,” Patt. Recogn., vol. 37, no. 12, pp. 2281–2291, 2004.