"... In PAGE 9: ... To train these models, we labeled a data set containing 20 minute clean speechfromTom Brokaw and 50 minute non-target audio data, including speech, environmentalsound,andmusic. Table1 provides the detailed results on face detection on the seven testing programs. The second column of the table gives the total number of keyframes for each program.... In PAGE 10: ... In our experiments, wesetup the thresholds so that the false alarm rate can be kept minimum during both face detection and anchor detection. Computed from the results in Table1 , the statistics yielded are: detection accuracy - 72%;; false rejection rate - 28%, and false acceptance rate - 0%. Examining the falsely rejected anchor frames, it was found that they fall into mostly two categories: poor qualityofanchor facial color (due to fade in/out, they are missed during face detection) and side views of the anchor (when the rotation is severe, the corresponding feature block does not possess the similar visual features as the ones from frontal views).... ..."

"... In PAGE 8: ... The performance results for both the face detection and the object tracking algorithms were obtained after visually analysing the detection and tracking results on each single frame and, using the criterion defined above, counting the number of correct and false locations and the number of missed faces. Table2 indicates the detection (in I-frames) and the tracking results (in all frames) for each of the two video sequences. As it can be seen, the face detection algorithm achieves a very high recall value of 93% and a slightly lower precision value of 82%.... ..."

"... In PAGE 5: ... For face recognition results, it was counted successful if the module identified the person correctly disregarding the facing angle. A summary of the results is compared in Table1 . With 25 eigenfaces, a 74%... ..."

"... In PAGE 4: ... Face databases from Leiden University, CMU, and MIT were used for testing. The size of each test set is given in Table1 . The 19th century database is composed of portrait photos as in Figure 3.... ..."

"... In PAGE 36: ... It is tested with Eye Analogue Detection (EAD) and without Eye Analogue Detection as described in [9]. Algorithm is tested with 1024 images containing various face ratios and the results are shown in Table1 . It is inferred that by employing Eye Analogue Detection the accuracy of the Face Detection module is improved as compared without Eye-Analog Detection, where Non-face Rejection Ratio (NRR) of the algorithm is very low, which results in the projection of non-feature points with the ... In PAGE 45: ... The results are validated for classification rate in percentage against the number of training samples. Table1 0: Classification rate using BPN with Individual Recognizer 10 20 30 40 50 60 70 80 90 No. of Training Samples (%) Classification rate in % LDA 63.... In PAGE 45: ... It is inferred that the recognition rate increases as the increase in the training set. Table1 1: Recognition rate and processing time for BPN network Training Set (%) Testing Set (%) Time For Training (Minutes: Seconds) Time For Testing (Minutes: Seconds) Recognition Rate (%) 90 10 57:24 02:00 97.6 80 20 49:36 09:18 93.... In PAGE 45: ....6.2. Individual Recognizers Validation with Kohonen network The classification rate is evaluated for the individual recognizers using Kohonen network. Table1 2: Classification rate using Kohonen network with Individual Recognizers 10 20 30 40 50 60 70 80 90 No. of Training Samples (%) Classification rate in % LDA 59.... In PAGE 45: ...6.3 70.5 75.3 80.3 83.3 86.7 90.1 92.2 93.6 From the validation results shown from Table1 2, for LBA, KDDA and LDA with Kohonen Self Organizing network, it is inferred that LBA is proved to be effective for higher classification rate. For different factors of training and testing sets, the processing time and the recognition rate are tabulated in Table 13 for Kohonen network and shows an increase in the recognition rate for a higher training set.... In PAGE 46: ...46 Table1 3: Recognition rate and processing time for Kohonen network Training Set (%) Testing Set (%) Time For Training (Minutes: Seconds) Time For Testing (Minutes: Seconds) Recognition Rate (%) 90 10 40:12 02:24 95.6 80 20 38:24 08:30 91.... In PAGE 46: ...4 are set for BPN. Table1 4: The classification rate of the hybrid classifier using BPN network 10 20 30 40 50 60 70 80 90 No. Of Training Samples (%) Classification rate in % LDA+KDDA 63.... In PAGE 47: ....Umamaheswari, S. Sumathi, S.N. Sivanandam and T. Ponson Table1 5: The classification rate of the hybrid classifier using Kohonen network 10 20 30 40 50 60 70 80 90 No. Of Training Samples (%) Classification rate in % LDA+KDDA 63.... In PAGE 47: ... Thus it is concluded that hybrid recognizer gives an efficient face detection module and high performance neural networks builds a reliable and adaptable face recognition system with high fault tolerance. Table1 6: Comparative results of BPN and Kohonen network Type of Neural Network BPN KOHONEN Training Set (%) Testing Set (%) Time For Training (Minutes: Seconds) Time For Testing (Minutes: Seconds) Recognition Rate (%) Time For Training (Minutes: Seconds) Time For Testing (Minutes: Seconds) Recognition Rate (%) 90 10 57:24 02:00 97.6 40:12 02:24 95.... In PAGE 70: ...g st(A) is an event occurring when the activity A is started. Table1 recapitulates some of cross references between events ... In PAGE 80: ... Because the risk of supply interruption is described by a single attribute, no weights are applied to its normalised rating. Table1 presents the set of input data for the model and Table 2 shows the corresponding output. All of the numerical cost figures in table 1 are measured in thousands of pounds.... In PAGE 93: ....1. Real Data Sets We experimented with different data sets, we give a brief description of the datasets used in our algorithm evaluation. The following Table1 shows some characteristics of the datasets. The real data used in the experiments were taken from http://www.... ..."

"... In PAGE 4: ... This detection test implies that a face exits when all its features are located. This way, the number of false-positive errors (see Table1 ) is very small. But many existing faces are difficult to be located, so we can also consider that a face is detected when a candidate region is found.... ..."