Figure 4-4 - uploaded by Tomaso A. Poggio
Content may be subject to copyright.
Illustration of match/nonmatch object pairs for Experiment 1. The top shows a pair a distance d = 0 :6 apart in morph space while the lower pair is separated by d = 0 :4.
Source publication
![Figure 3-1: (a) Cartoon of the Poggio and Edelman model [68] of...](profile/Tomaso-Poggio-2/publication/2851007/figure/fig1/AS:340642740883457@1458226966317/1-a-Cartoon-of-the-Poggio-and-Edelman-model-68-of-view-based-object-recognition_Q320.jpg)
this paper was indeed that a MAX-like operation may represent the cortical equivalent of the "window of analysis" in machine vision to scan through and select input data. Unlike a centrally controlled sequential scanning operation, a mechanism like the MAX operation that locally and automatically selects a relevant subset of inputs seems biological...
Citations
... The simple cells are arranged in cell-planes and each of these planes reacts to a specific stimulus in different positions, after learning each cell-plane becomes independent. HMAX [4,15,5] also builds on the classical hypothesis of Hubel and Wiesel. A key difference between HMAX and Neocognitron is the complex cells responses which on the first is the maximum of the afferent responses and on the second the sum [16] or squared sum [17] of the afferent responses. ...
Image recognition problems are usually difficult to solve using raw pixel data. To improve the recognition it is often needed some form of feature extraction to represent the data in a feature space. We use the output of a biologically inspired model for visual recognition as a feature space. The output of the model is a binary code which is used to train a linear classifier for recognizing handwritten digits using the MNIST and USPS datasets. We evaluate the robustness of the approach to a variable number of training samples and compare its performance on these popular datasets to other published results. We achieve competitive error rates on both datasets while greatly improving relatively to related networks using a linear classifier.
The paper presents a method of image recognition, which is inspired by research in visual cortex. The architecture of our model called CaNN is similar to the one proposed in neocognitron, LeNet or HMAX networks. It is composed of many consecutive layers with various number of planes (receptive fields). Units in the corresponding positions of the planes in one layer receive input from the same region of the precedent layer. Each plane is sensitive to one pattern. The method assumes that the pattern recognition is based on edges, which are found in the input image using Canny detector. Then, the image is processed by the network. The novelty of our method lies in the way of information processing in each layer and an application of clustering module in the last layer where the patterns are recognized. The transformations performed by the CaNN model find the own representation of the training patterns. The method is evaluated in the experimental way. The results are presented.
