J. J. Hull. A database for handwritten text recognition research. IEEE Transactions on Pattern Analysis and Machine Intelligence, 16(5):550--554, May 1994.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....classi er. The database contains isolated characters extracted from handwritten words coming from two sources: the rst is the database of handwritten words produced by the Center of Excellence for Document Analysis and Recognition (CEDAR) at the State University of New York (SUNY) at Bu alo [2].The second is a database of handwritten postal addresses digitalized by the USPS (United States Postal Service) Word images from which the characters were extracted were all preprocessed according to a scheme that includes morphologic ltering, deslanting and deskewing [9] The character ....

....scheme that includes morphologic ltering, deslanting and deskewing [9] The character database contains both uppercase and lowercase letters. Figure 2 shows the percentage distribution of letters in the test set. Since the data is extracted from a database collected by USPS in a real postal plant [2], our database distribution re ects the prior distribution of that site. For this reason some letters are less represented than others or almost absent. Clustering performed with Self Organizing Maps (SOM) 10] and Neural Gas (NG) 12] showed that, for some letters, the vectors corresponding to ....

J.J.Hull, A Database for Handwritten Text Recognition Research, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 16, No. 5, 550-554, 1994.

....only. For example the NIST database [2] is based on a small vocabulary. In the database described in [15] only a single writer is represented. Other databases have been built for speci c tasks like the recognition of handwritten legal amounts [7, 4] or postal addresses as in the CEDAR database [5, 3, 9]. More recently a database containing essays written by students was presented in [1] In order to train and test systems which are able to recognize unconstrained English texts as described in [8] or [11, 12] the IAM database has been built at our institute. This database is available upon ....

J.J. Hull. A database for handwritten text recognition research. IEEE Trans. on Pattern Analysis and Machine Intelligence, 16(5):550-554, May 1994.

....2; u = 184; 0:43. Both s are from the MST heuristic. Simple ML classification is used. Here, obviously kNN would fail to follow the structure of data. For a real world example we test label propagation on a handwritten digits dataset, originally from the Cedar Buffalo binary digits database [4]. The digits were preprocessed to reduce the size of each image down to 16 16 by down sampling and Gaussian smoothing, with pixel values ranging from 0 to 255 [5] We use digits 1 , 2 and 3 in our experiment as three classes, with 1100 images in each class. Each image is represented by a 256 ....

Jonathan J. Hull. A database for handwritten text recognition research. IEEE Transactions on Pattern Analysis and Machine Intelligence, 16(5), 1994.

....2 2.5 3 3.5 1 2 1 0 2 0.5 1.5 2.5 3.5 1 2 1 0.5 1.5 2.5 3.5 (c) Label propagation Figure 2: The Springs dataset. For a real world example, we test label propagation on a handwritten digits dataset. The original dataset was from the Cedar Bu alo binary digits database [Hul94] The digits were then preprocessed to reduce the size of each digit image down to 16x16 by down sampling and Gaussian smoothing. This interpolation thus created gray scale with pixel values range from 0 to 255 [CBD 90] We use digits 1, 2 and 3 in our experiment as three classes. Each class ....

Jonathan J. Hull. A database for handwritten text recognition research. IEEE Transactions on Pattern Analysis and Machine Intelligence, 16(5), 1994.

....proceedings of NIPS conference papers. Both of these datasets are likely to have intrinsic structure in many fewer dimensions than their raw dimensionalities: 256 for the handwritten digits and 13679 for the author word counts. To begin, we used a set of 3000 digit bitmaps from the UPS database[7] with 600 examples from each of the five classes 0,1,2,3,4. The variance of the Gaussian around each point in the 256 dimensional raw pixel image space was set to achieve a perplexity of 15 in the distribution over high dimensional neighbors. SNE was initialized by putting all the y i in random ....

J. J. Hull. A database for handwritten text recognition research. IEEE Transaction on Pattern Analysis and Machine Intelligence, 16(5):550--554, May 1994.

....proceedings of the NIPS conference papers. Both of these datasets are likely to have intrinsic structure in many fewer dimensions than their raw dimensionalities: 256 for the handwritten digits and 13679 for the word counts. To begin, we used a set of 3000 digit images from the UPS database[7] with equal numbers from each of the five classes 0,1,2,3,4. The variance of the Gaussian around each point in the 256 dimensional image space was set to achieve a perplexity of 15 in the distribution over high dimensional neighbors. SNE was initialized by putting all the digits in random ....

J. J. Hull. A database for handwritten text recognition research. IEEE Transaction on Pattern Analysis and Machine Intelligence, 16(5):550--554, May 1994.

....results will allow us to confirm the validity of our approach. 6 Experiments This section describes the characteristics of the used database and presents the results obtained with the experiments held with the Group 01 and 02 models. 6. 1 Databases There exists several international databases [6,7] of handwritten checks. However, these databases do not deal with the Portuguese language. Owing to the difficulties of obtaining databases with real document checks through national bank institutions, the creation of a bank check laboratory database was chosen. The acquisition of the images was ....

J.J. Hull, A database for handwritten text recognition research, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol.16, No. 5, 1794, 550--554.

....have better acceleration performance than the proposed adaptive leaming rate. 1V. SIMULATION A handwritten digit recognition problem was used to verify the effectiveness of the adaptive leaming rate and limited error signals. A total of 18,468 handwritten digitized images from the CEDAR database [24] were used for training after size normalization. A digit image consisted of 12x 12 pixels and each pixel took on integer values from 0 to 15. Figure 2 shows some examples of digit images. The MLP consisted of 144 inputs, 30 hidden nodes, and 10 output nodes. Initialized weights were drawn at ....

J. J. Hull, "A Database for Handwritten Text Recognition Research, " IEEE Trans. Pat. Ana. Mach. Int., Vol. 16, No. 5, May 1994, pp. 550-554.

....number in the case was 90. Since the number is between 72 and 126, the intrinsics dimension, estimated by TRN, is between 6 and 7. As in the rst benchmark, ID estimated by TRN based method is lower than the one computed by GP. 4 Characters were selected by words belonging to CEDAR database [11]. npl.tex; 15 01 2001; 13:12; p.11 12 5. Conclusions We presented an empirical approach based on Grassberger Procaccia s algorithm in order to estimate the intrinsic dimensionality of data. The approach was applied to two di erent benchmarks and the results obtained were compared to those given ....

J.J.Hull, \A database for handwritten text recognition research", IEEE Trans. on Pattern Analysis and Machine Intelligence, Vol. 16, No.5, May

....relative responsibilities for the remaining beads is guaranteed to improve F . VI. Results on isolated digits The performance of the elastic net in recognizing isolated digits has been tested on data from the CEDAR CDROM 1 database of Cities, States, ZIP Codes, Digits, and Alphabetic Characters [44]. The br training set of binary segmented digits was subdivided into 3 training sets of size 2000, 7000 and 2000 respectively. A validation set of 2000 examples was also generated from the br training set to allow as to investigate different configurations of the post processing neural network. ....

....net on the third set. The CEDAR database also includes 2 test sets. The goodbs (2213 images) set is a subset of the bs (2711 images) set containing only well segmented digits. It is interesting to note that br training data were segmented with the same diligence as the goodbs test data [44]. Validation goodbs bs Set test set test set Full Covariance 2.00 1.85 3.58 Matrix Diagonal Covariance 1.75 1.53 3.43 Matrix Mixture of 1.00 1.50 3.14 local models Table 1: Percentage of images incorrectly classified by the elastic net with no rejections. In comparing our error rates with ....

J. J Hull, "A database for handwritten text recognition research", IEEE Trans. Pattern Anal. Machine Intell., vol. 16, no. 5, pp. 550--554, 1994.

....of a person walking outdoors. The head has different poses and appears at different positions in the field of view. In addition, the background is highly cluttered and there is variation in lighting conditions. Fig. 3 shows preprocessed greyscale images of handwritten digits from postal envelopes [2]. Unlike the microscope images described above, in this case the boundaries of the digits on the envelopes were more easily identifiable, so the digits are normalized for horizontal and vertical scale and translation before sampling the 8 8 pixel images shown in the figure. However, the digits ....

J. J. Hull, "A database for handwritten text recognition research," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 16, no. 5, pp. 550--554, 1994.

....can also be incorporated directly into the GVQ algorithm by using s i 2 f0; 1; Kg. An example using two three valued features is shown in figure 3 b. 3 Results 3. 1 Handwritten digits We randomly selected 400 training images of handwritten threes and fives from the CEDAR CDROM 1 database [6]. Since the original images contain different numbers of pixels, we rescaled all images to 20 Theta 20 pixels. In contrast to [1] and [7] there was no need to reduce the image size further since the algorithm convergences quickly scaling only linearly with the number of input dimensions. We ....

J. J. Hull. A database for handwritten text recognition research. IEEE Transactions on Pattern Analysis and Machine Intelligence, 16(5):550--554, 1997.

....of 10 factor analyzers and a 10 cluster TMG trained on all 2000 digits. In each case, the best of 10 experiments was selected. 4.3 Modeling Handwritten Digits. We performed both supervised and unsupervised learning experiments on 8 8 greyscale versions of 2000 digits from the CEDAR CDROM (Hull, 1994). Although the preprocessed images t snugly in the 8 8 window, there is wide variation in writing angle (e.g. the vertical stroke of the 7 is at di erent angles) So, we produced a set of 29 shearing translation transformations (see the top row of Fig. 4b) to use in transformed density ....

J. J. Hull 1994. A database for handwritten text recognition research. IEEE Trans. on Pattern Analysis and Machine Intelligence 16:5, 550-554.

No context found.

J. J. Hull. A database for handwritten text recognition research. IEEE Transactions on Pattern Analysis and Machine Intelligence, 16(5):550--554, May 1994.

No context found.

Hull, J. J. (1994). A database for handwritten text recog- nition research. IEEE Transactions on Pattern Analysis and Machine Intelligence, 16.

No context found.

J. J. Hull. A database for handwritten text recognition research. IEEE Transaction on Pattern Analysis and Machine Intelligence, 16(5):550--554, May 1994.

No context found.

Hull, J. J. (1994). A database for handwritten text recog- nition research. IEEE Transactions on Pattern Analysis and Machine Intelligence, 16.

No context found.

Hull, J. J. (1994). A database for handwritten text recognition research. IEEE Transactions on Pattern Analysis and Machine Intelligence, 16(5):550--554.

No context found.

J. J Hull, "A database for handwritten text recognition research", IEEE Transactions Pattern Analysis and Machine Intellegince,vol. 16, no. 5, pp. 550--554, 1994.

No context found.

J. J. Hull. A database for handwritten text recognition research. IEEE Transactions on Pattern Analysis and Machine Intelligence, 16(5):550--554, May 1994.

No context found.

J. J. Hull. A database for handwritten text recognition research. IEEE Transactions on Pattern Analysis and Machine Intelligence, 16(5):550--554, May 1994.

No context found.

Hull, JJ (1994). A database for handwritten text recognition research. IEEE Transactions on Pattern Analysis and Machine Intelligence, 16, 550-554.

No context found.

J.J. Hull. A database for handwritten text recognition research. IEEE Transactions on Pattern Analysis and Machine Intelligence, 16(5):550-554, May 1994.

No context found.

J.J. Hull. A database for handwritten text recognition research. IEEE Transactions on Pattern Analysis and Machine Intelligence, 16(5):550-554, 1994.

No context found.

Hull, J.J. A database for handwritten text recognition research. IEEE Transactions on Pattern Analysis and Machine Intelligence, 16(5):550-554, 1997.

First 50 documents  Next 50

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC