The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
The Interna 
1086 
in the high-dimensional space and the image manifold is sampled 
at regular intervals of pose angle to make the training images. In 
the first experiment, images separated by 50° in pose angle were 
chosen to construct the representative non Gaussians space of the 
images. That is to say, there are 7 training images for each object, — jj 
making a total of 140 training images. The training views of the 
first object are shown in Fig 4 Table I Ni 
3.1. Results with Pose Angle Sampling at Every 50° 
The image presented to the IPCA network as described in the 
methodology. The recognition is achieved by finding th e 
minimum distance between the coefficients of a test image and 
the training images. The vector of the image in the training set 
that is nearest to the test image non Gaussian vector is chosen as 
the recognized image. 
3.2. Result 
The previous ex 
every 25°. This j 
images as the tes 
Training images are sampled by 50° in pose angle. That is to say, 
there are 6 training images for each object, making a total of 140 
training images. The images that were not in the training set were 
considered as test images, thus making a total of 1300 test images. 
In the following two experiments, number of non Gaussian 
vectors q is turned parameters q= {20, 25,30,35,40, 45, 50} and 
the Table II shows the results and the Fig 5 represents the test 
applied to that object 
J< 
_P 
II 
Table IIN 
From the a 
better resi 
increasing 
that the pe 
the Increm 
vectors i 
vectors are 
(a) 
Fig 5: Object Extracting Using IPCA-ICA (50°) 
(a) Input image (b) Object is mixed with some unknown mixtures 
(c) Object Extracting Using IPCA-ICA (d) Duck with Orient 150°