Multilayer Architectures for Facial Action Unit Recognition.

In expression recognition and many other computer vision applications, the recognition performance is greatly improved by adding a layer of nonlinear texture filters between the raw input pixels and the classifier. The function of this layer is typically known as feature extraction. Popular filter types for this layer are Gabor energy filters (GEFs) and local binary patterns (LBPs). Recent work [1] suggests that adding a second layer of nonlinear filters on top of the first layer may be beneficial. However, it is unclear what is the best architecture of layers and selection of filters. In this paper, we present a thorough empirical analysis of the performance of single-layer and dual-layer texture-based approaches for action unit recognition. For the single hidden layer case, GEFs perform consistently better than LBPs, which may be due to their robustness to jitter and illumination noise as well as to their ability to encode texture at multiple resolutions. For dual-layer case, we confirm that, while small, the benefit of adding this second layer is reliable and consistent across data sets. Interestingly for this second layer, LBPs appear to perform better than GEFs.

Human and computer recognition of facial expressions of emotion.

Neuropsychological and neuroimaging evidence suggests that the human brain contains facial expression recognition detectors specialized for specific discrete emotions. However, some human behavioral data suggest that humans recognize expressions as similar and not discrete entities. This latter observation has been taken to indicate that internal representations of facial expressions may be best characterized as varying along continuous underlying dimensions. To examine the potential compatibility of these two views, the present study compared human and support vector machine (SVM) facial expression recognition performance. Separate SVMs were trained to develop fully automatic optimal recognition of one of six basic emotional expressions in real-time with no explicit training on expression similarity. Performance revealed high recognition accuracy for expression prototypes. Without explicit training of similarity detection, magnitude of activation across each emotion-specific SVM captured human judgments of expression similarity. This evidence suggests that combinations of expert classifiers from separate internal neural representations result in similarity judgments between expressions, supporting the appearance of a continuous underlying dimensionality. Further, these data suggest similarity in expression meaning is supported by superficial similarities in expression appearance.

Face recognition by independent component analysis.

A number of current face recognition algorithms use face representations found by unsupervised statistical methods. Typically these methods find a set of basis images and represent faces as a linear combination of those images. Principal component analysis (PCA) is a popular example of such methods. The basis images found by PCA depend only on pairwise relationships between pixels in the image database. In a task such as face recognition, in which important information may be contained in the high-order relationships among pixels, it seems reasonable to expect that better basis images may be found by methods sensitive to these high-order statistics. Independent component analysis (ICA), a generalization of PCA, is one such method. We used a version of ICA derived from the principle of optimal information transfer through sigmoidal neurons. ICA was performed on face images in the FERET database under two different architectures, one which treated the images as random variables and the pixels as outcomes, and a second which treated the pixels as random variables and the images as outcomes. The first architecture found spatially local basis images for the faces. The second architecture produced a factorial face code. Both ICA representations were superior to representations based on PCA for recognizing faces across days and changes in expression. A classifier that combined the two ICA representations gave the best performance.

Alteration in expression of the rat mitochondrial ATPase 6 gene during Pneumocystis carinii infection.

BACKGROUND: Pneumocystis carinii causes pneumonia in immunocompromised patients with a high morbidity and mortality rate, but the interaction between this organism and the host cell is not well understood. The purpose of this research was to study the response of host cells to P. carinii infection on a molecular level. RESULTS: The technique of mRNA differential display was used to detect genes whose expression may be affected by P. carinii infection. The nucleotide sequence of one differentially displayed DNA fragment was found to be identical to that of the rat mitochondrial ATPase 6 gene, which is a subunit of the F0F1-ATP synthase complex. A four-fold increase in expression of this gene was verified by Northern blot analysis of total RNA extracted from P. carinii-infected rat lung versus that from mock-infected rat lung. Localization of the cells containing ATPase 6 mRNA was accomplished by in situ hybridization. In sections of non-infected rat lung, these cells were found lining the distal parts of the respiratory tree and in apical areas of the alveoli. Histological location of these cells suggested that they were Clara cells and type II pneumocytes. This hypothesis was confirmed by co-localizing the mRNAs for ATPase 6 and surfactant protein B (SP-B) to the same cells by two-color fluorescent in situ hybridization. CONCLUSIONS: The ATPase 6 gene is over expressed during P. carinii infection, and type II pneumocytes and Clara cells are the cell types responsible for this over-expression.

Internal transcribed spacer regions of rRNA genes of Pneumocystis carinii from monkeys.

Analysis of sequence variations among isolates of Pneumocystis carinii f. sp. macacae from 14 Indian rhesus monkeys (Macaca mulatta) at the internal transcribed spacer (ITS) regions of the nuclear rRNA gene was undertaken. Like those from P. carinii f. sp. hominis, the ITS sequences from various P. carinii f. sp. macacae isolates were not identical. Two major types of sequences were found. One type of sequence was shared by 13 isolates. These 13 sequences were homologous but not identical. Variations were found at 13 of the 180 positions in the ITS1 region and 28 of the 221 positions in the ITS2 region. These sequence variations were not random but exhibited definite patterns when the sequences were aligned. According to this sequence variation, ITS1 sequences were classified into three types and ITS2 sequences were classified into five types. The remaining specimen had ITS1 and ITS2 sequences substantially different from the others. Although some specimens had the same ITS1 or ITS2 sequence, all 14 samples exhibited a unique whole ITS sequence (ITS1 plus ITS2). The 5.8S rRNA gene sequences were also analyzed, and only two types of sequences that differ by only one base were found. Unlike P. carinii f. sp. hominis infections in humans, none of the monkey lung specimens examined in this study were found to be infected by more than one type of P. carinii f. sp. macacae. These results offer insights into the genetic differences between P. carinii organisms which infect distinct species.

Activities and conformational fitting of 1,4-naphthoquinone derivatives and other cyclic 1,4-diones tested in vitro against Pneumocystis carinii.

Atovaquone is a chemotherapeutic agent used to treat pneumonia caused by Pneumocystis carinii in some immunocompromised patients. A set of cyclic 1,4-diones were tested in vitro for ability to inhibit growth of P. carinii, including 22 variously substituted 1,4-naphthoquinones, one bis-1,4-naphthoquinone, and three other quinones. For comparison, the antipneumocystic primaquine and its 5-hydroxy-6-desmethyl metabolite were also tested. At 1.0 microg/ml, seven compounds inhibited growth by at least 39%, with atovaquone at 92%; of these seven, five are 2-hydroxy-1,4-naphthoquinones, while one is a 2-chloro- and another is a 2-methyl-1,4-naphthoquinone. At 0.1 microg/ml, however, the most active compound tested was the primaquine metabolite, which inhibited growth by more than 42% at this concentration. To ascertain a structure-activity relationship, all 1,4-naphthoquinones were compared conformationally by means of computer-based molecular modeling (Spartan) incorporating the Sybyl force field. Without exception, for all 21 monomers tested, the substituent at position 3 of the 1,4-naphthoquinone favored activity most strongly when it simultaneously occupied (i) space centered at about 3 A from position 3, without projecting steric bulk from the area encompassed by atovaquone's cyclohexyl ring, and (ii) roughly planar space at about 7.3 A from position 3, without projecting steric bulk perpendicularly. This structure-activity relationship may prove useful in the rational design of better antipneumocystis agents.

Characterization of the gene encoding a histidine and aspartic acid-rich protein from Pneumocystis carinii.

A cDNA clone derived from Pneumocystis carinii contained an unusual sequence (GTGATG)2(ATGGTG)4(ATG)4 and many GAT repeats. It was found to encode a histidine and aspartic acid-rich protein (HARP). The complete cDNA contained an 888-bp open reading frame encoding a putative protein of 32.6 kDa. The deduced HARP protein contained 39 aspartic acid and 22 histidine residues. The genomic copy of the HARP gene (1203 bp in length) was found to contain 3 small introns of 46, 44, and 38 bp, respectively. HARP was predicted by computer programs to be a plasma membrane protein with nickel-binding activity.

The orientation of DNA in an archaeal transcription initiation complex.

RNA polymerase from the hyperthermophile archaeon Pyrococcus furiosus (Pfu) forms specific and transcriptionally active complexes with its conjugate transcription factors TBP (the archaeal TATA binding protein homolog) and TFB (the archaeal homolog of eukaryotic RNA polymerase II and III transcription factors TFIIB and Brf) at the Pfu glutamate dehydrogenase promoter. A photochemical crosslinking method was used to map the vicinity of the catalytic subunits of Pfu RNA polymerase to DNA locations distributed along the polymerase-promoter interface. The largest component of this archaeal polymerase is split into two subunits, A' and A", whose relatively sharp boundary of DNA crosslinking (probed on the transcribed strand) is centered five to six base pairs downstream of the transcriptional start site. A strong argument based on this information, on the well-defined homology between the core bacterial, archaeal and eukaryotic RNA polymerase subunits, and on the recently determined structure of a bacterial RNA polymerase specifies the directionality of DNA in the archaeal transcription complex and its trajectory downstream of the transcriptional start site.

Down-regulation of GATA-2 transcription during Pneumocystis carinii infection.

Differences in gene expression between Pneumocystis carinii-infected and noninfected rats were examined. Total RNA was isolated from homogenized rat lungs and then subjected to differential display with combinations of oligo(dT) and various arbitrary PCR primers. Approximately 50 differentially expressed bands were observed. Several of these DNA bands were isolated, reamplified, and cloned. The cloned DNA fragments were used as probes to perform Northern hybridization on RNA from P. carinii-infected and noninfected rat lungs. One clone was found to react with a 3-kb mRNA from noninfected but not from P. carinii-infected rat lung, suggesting that the gene represented by this clone was down-regulated during P. carinii infection. The nucleotide sequence of this clone was determined and found to be 97% homologous to the mouse GATA-2 transcription factor. In situ hybridization using RNA probes derived from this clone revealed that alveolar macrophages, resident lung monocytes, and bronchial epithelial cells express the GATA-2 gene in the lung.

Regulation of rRNA transcription is remarkably robust: FIS compensates for altered nucleoside triphosphate sensing by mutant RNA polymerases at Escherichia coli rrn P1 promoters.

We recently identified Escherichia coli RNA polymerase (RNAP) mutants (RNAP beta' Delta215-220 and beta RH454) that form extremely unstable complexes with rRNA P1 (rrn P1) core promoters. The mutant RNAPs reduce transcription and alter growth rate-dependent regulation of rrn P1 core promoters, because the mutant RNAPs require higher concentrations of the initiating nucleoside triphosphate (NTP) for efficient transcription from these promoters than are present in vivo. Nevertheless, the mutants grow almost as well as wild-type cells, suggesting that rRNA synthesis is not greatly perturbed. We report here that the rrn transcription factor FIS activates the mutant RNAPs more strongly than wild-type RNAP, thereby compensating for the altered properties of the mutant RNAPs. FIS activates the mutant RNAPs, at least in part, by reducing the apparent K(ATP) for the initiating NTP. This and other results suggest that FIS affects a step in transcription initiation after closed-complex formation in addition to its stimulatory effect on initial RNAP binding. FIS and NTP levels increase with growth rate, suggesting that changing FIS concentrations, in conjunction with changing NTP concentrations, are responsible for growth rate-dependent regulation of rrn P1 transcription in the mutant strains. These results provide a dramatic demonstration of the interplay between regulatory mechanisms in rRNA transcription.

Typing of Histoplasma capsulatum isolates based on nucleotide sequence variation in the internal transcribed spacer regions of rRNA genes.

The nucleotide sequences of internal transcribed spacer (ITS) regions of rRNA genes of 24 isolates of Histoplasma capsulatum were examined. The results indicate that the sequences of ITS regions in different isolates are not identical. Sequence variations were found at 20 positions in the 496 bp that were sequenced. Ten different sequence patterns, designated types A through H, were observed when the sequences from the 24 isolates were aligned. Twelve isolates from Indianapolis were classified into four different types. Two isolates from New York belonged to type G. Three isolates from different cities were type F. The remaining six isolates were of different types.

Diagnosis of Pneumocystis carinii pneumonia in human immunodeficiency virus-infected patients with polymerase chain reaction: a blinded comparison to standard methods.

Pneumocystis carinii pneumonia (PCP) is an important cause of morbidity and death among persons with human immunodeficiency virus (HIV) infection. Polymerase chain reaction (PCR) analysis of respiratory specimens has been investigated as a rapid diagnostic method. We have previously reported on the utility of this technique for diagnosing PCP in HIV-infected patients. In this report we evaluate PCR used in a blinded study design to avoid biases inherent to retrospective and nonblinded studies. The diagnosis of PCP was established on the basis of clinical findings and morphological studies of bronchoalveolar lavage (BAL) and/or lung biopsy specimens before PCR testing. PCR was performed without knowledge of the diagnosis. PCR results were graded from "negative" to 3+ on the basis of intensity of the banding pattern. Forty-seven patients were enrolled in the study, including 18 with proven PCP and 29 with other conditions. PCR was positive at grade 1 or higher for all 18 patients with PCP (100% sensitivity), at grade 2 or higher for 13 patients (72.2% sensitivity), and at grade 1 or higher for 4 of the 29 control patients (specificity of 86.2%). If a grade 2 or higher was required for diagnosis, the specificity improved to 100%. Results were reproducible with testing of a second aliquot for 46 of 47 patients (97.8%). Our findings confirm that PCR is a sensitive and reproducible test for detection of P. carinii in BAL specimens. Problems with false-positive results for control patients, however, limit the applicability of this method.

Measuring facial expressions by computer image analysis.

Facial expressions provide an important behavioral measure for the study of emotion, cognitive processes, and social interaction. The Facial Action Coding System (Ekman & Friesen, 1978) is an objective method for quantifying facial movement in terms of component actions. We applied computer image analysis to the problem of automatically detecting facial actions in sequences of images. Three approaches were compared: holistic spatial analysis, explicit measurement of features such as wrinkles, and estimation of motion flow fields. The three methods were combined in a hybrid system that classified six upper facial actions with 91% accuracy. The hybrid system outperformed human nonexperts on this task and performed as well as highly trained experts. An automated system would make facial expression measurement more widely accessible as a research tool in behavioral science and investigations of the neural substrates of emotion.

Intranasal immunization confers protection against murine Pneumocystis carinii lung infection.

To evaluate the feasibility of mucosal immunization against Pneumocystis carinii (Pc) experimental infection, female BALB/c mice were intranasally immunized three times with soluble Pc antigens plus cholera toxin fraction B (Pc-CTB); control groups received either Pc antigen, CTB, or phosphate-buffered saline (PBS) alone. Two weeks after the last immunization, five animals from each group were sacrificed, and cellular and humoral immune responses were evaluated. The remaining five mice were CD4 depleted using a monoclonal antibody against mouse CD4 and inoculated with viable Pc. Significantly higher specific lymphoproliferative responses from tracheobronchial lymph node cells, immunoglobulin M (IgM) and IgG antibody levels in serum, and bronchoalveolar lavage (BAL)-derived IgA antibody concentrations were observed in the Pc-CTB group of mice relative to control groups (P < 0.01). Five weeks after challenge, no Pc organisms were observed in the lung smears of the Pc-CTB group, while the animals receiving antigen, adjuvant, or PBS had progressively higher numbers of Pc microorganisms. By Western blot analysis, a strongly reactive 55- to 60-kDa antigen was recognized by BAL IgA and by serum IgG. In summary, mucosal immunization elicited specific cellular and humoral immune responses and protected against Pc lung infection after immunosuppression.

Learning viewpoint-invariant face representations from visual experience in an attractor network.

In natural visual experience, different views of an object or face tend to appear in close temporal proximity as an animal manipulates the object or navigates around it, or as a face changes expression or pose. A set of simulations is presented which demonstrate how viewpoint-invariant representations of faces can be developed from visual experience by capturing the temporal relationships among the input patterns. The simulations explored the interaction of temporal smoothing of activity signals with Hebbian learning in both a feedforward layer and a second, recurrent layer of a network. The feedforward connections were trained by competitive Hebbian learning with temporal smoothing of the post-synaptic unit activities. The recurrent layer was a generalization of a Hopfield network with a low-pass temporal filter on all unit activities. The combination of basic Hebbian learning with temporal smoothing of unit activities produced an attractor network learning rule that associated temporally proximal input patterns into basins of attraction. These two mechanisms were demonstrated in a model that took grey-level images of faces as input. Following training on image sequences of faces as they changed pose, multiple views of a given face fell into the same basin of attraction, and the system acquired representations of faces that were approximately viewpoint-invariant.