With the new Hamburg astrometric measuring machine, large sets of plates can be digitized very quickly with submicrometer accuracy. In particular about 2000 plates of the AGK2-catalog, mean epoch 1930, can be remeasured now for the first time to their limiting magnitude, about B = 11. The new AGK2-data therefore will cover practically the whole AC-catalog and TYCHO-stars in the northern hemisphere.

All plates will be reduced using the HIPPARCOS results as the reference frame when available in 1996. The new AGK2-data will provide a unique data base for the determination of high accuracy proper motions (about 2 mas/yr) of all TYCHO stars in the northern hemisphere. Furthermore, for the first time a dense reference frame for a final reduction of the Astrographic Catalog (AC) and the large deep sky surveys will be generated by this catalog. The inferior situation in the southern hemisphere will be addressed briefly.

Over the last 18 years or so a large-scale monitoring programme has been undertaken with the UK 1.2 m Schmidt telescope in field 287 at 21h 28m, −45 deg. Plates have been taken over all timescales from 1 hour to 17 years in red and blue passbands, and large numbers of plates have also been taken in other passbands to give deep coverage in U, B, V, R and I. The field is calibrated with some 200 CCD and photoelectric standards, and covers approximately 25 square degrees.

DIA (Digital Image Analyzer) is a new dedicated and low cost CCD line scanner, which has been invented at Sonneberg Observatory in collaboration with the Institute for Theoretical Astrophysics at Tübingen for scanning selected fields of the Sonneberg Plate Archive.

Several types of fast digitization devices consisting of CCD cameras and frame grabbers have appeared and been implemented in astronomical practice during the last decade. Nevertheless, various types of classical scanning devices are still used for precise digitization of astronomical images and spectrograms. Among them, a new type of microphotometer with a rotating slit (UMF) was adjusted and installed at the Ondřejov Observatory (Zicha et al. 1992). Technical parameters of the device are as follows:

Realization of solar corona polarimetry pertains to astrophysical centre-symmetrical problems. Usual ways of digitization are not optimum for this task. Coronal images obtained during solar eclipse at different angles of the polarizer should be digitized symmetrically in relation to the centre of the solar disk. Unfortunately, digitization of such centre-symmetrical images is usually performed by rectangular scanning with a firmly oriented digitization slit. Corresponding pixels in different images have to be found by methods of shifting the centres of the relevant images and a necessary rotation of the images. When the images are almost identical, or at least very similar each to other, the final fitting of the images can be tested by calculating the best correlation. As a rule, this is not the case for solar corona images obtained with different angle of the polarizer. In addition, after rotation of the digitized image, the digitisation rectangular networks become incompatible and thus some information is changed or even lost.

The extraction and characterisation of information from photographic emulsions is the subject of this review. Since the major Schmidt Telescope full sky surveys form the bulk of the photographic emulsions currently under intensive study, I will limit my remarks to them, although most would be applicable to any of the emulsions now in use. A listing of the major direct imaging surveys is given in the article by van Altena et al (1993).

We have developed a system, called SKICAT, for producing, managing and analyzing catalogs from the digitized POSS-II survey. The system classifies and matches catalogs from multiple, overlapping plate scans as well as CCD calibration sequences; and it can be used for the scientific analysis of the resulting catalogs. It incorporates a number of novel machine-learning and AI tools, including the star/galaxy classification using decision tree algorithms. This results in star/galaxy separation accurate to 90% or better down to BJ ∼ 21m, i.e. ∼ 1m above the plate limit The final catalog is expected to contain at least 5 × 107 galaxies and > 2 × 109 stars. We present preliminary results on galaxy counts from a test region near the NGP. We find a mild excess over the no-evolution models, smaller than previously found by the APM group. A search for z > 4 quasars and the two-point correlation analysis of this data set are in progress.

Today, large astronomical plates are digitised with fast scanners, leading to images with about 109 pixels. This amount of information permits astronomers to apply various computer vision techniques to get inventories of the objects on the plates.

Efficient data compression needs analyzing functions to recognise the local resolution of the signal. They are provided by the wavelet concept. The optimal wavelet (best information concentration) is defined by the image model of the application. For the most common images in astronomy the H-transform is optimal (in the sense of Karhunen-Loeve transform). The role of the H-transform in 2-dimensional processing is the same as the Haar-transform in 1-dimension, but it is not the ‘2-dimensional Haar-transform’ found in text books.

The automatic identification and characterization of star images has great value for the preliminary analysis and measurement of astrographic plates. Our group at Valencia Observatory is using a small 2-D stage and a CCD camera under computer control to perform systematic measurements of bright asteroid plates. We are also applying this method to the processing of astrographic plates with crowded stellar fields and non-stellar objects in collaboration with the Pulkovo Observatory.

A short overview is presented of a number of different discriminant analysis methods. An ‘uncertainty principle’ is presented in regard to the issue of user choice of appropriate method. The discriminant analysis methods described are then used in the important problem of feature selection. A hand-classified set of HST Guide Star plate data is used, with star/galaxy/fault classes.

The use of neural network pattern recognition techniques in the field of astronomy is reviewed. In assessing the quality of image recognition derived from this method particular attention is given to the problem of star/galaxy discrimination in large digital sky surveys. A two color survey of 9 fields of the first epoch Palomar Sky Survey, centered on the North Galactic Pole, has been performed with the Minnesota Automated Plate Scanner. A set of neural network image classifiers are used to automatically perform star/galaxy discrimination. We assess the efficiency of image classification and sample completeness through comparisons with a variety of independent studies of the NGP area.

We study the performance and limitations of the morphological classification method based on luminosity concentration and mean surface brightness. In particular, the effects of the different colour bands and of a finite seeing are investigated.

Artificial Neural Network techniques are applied to the classification of faint objects, detected in digital astronomical images, and a Bayesian classifier (the neural network classifier, NNC hereafter) is proposed. This classifier can be implemented using a feedforward multilayered neural network trained by the back-propagation procedure (Werbos 1974).

Some years ago a complex programme of studying the main meridional section of the Galaxy was started by astronomers of Kiev, Tartu, Abastumani and Vilnius Observatories with the aim of improving our knowledge of spatial and kinematic characteristics of stellar populations. Characteristic to the programme is the use of absolute proper motions of stars together with automated quantitative spectral classification for large stellar-statistical samples. The data are gathered in areas lying within 30° of the main meridional section of the Galaxy. To classify stars, objective prism stellar spectra of intermediate dispersion (166 å/mm at Hγ), obtained with the 70 cm meniscus telescope at the Abastumani Astrophysical Observatory, are used. The field diameter is 4° 50′, and the limiting photographical stellar magnitude is about 12m. Our system of automated quantitative spectral classification of F-K stars applies criteria evaluation and is mainly based on two software packages: the SDR package for spectrometric data reduction and the CTATEC-2 package determining the linear regression model used for classification (Malyuto & Shvelidze 1989; Malyuto, Pelt & Shvelidze 1993).

A spectral classification scheme has been developed for objective prism spectra of LMC stars, taken with the 1.2 m UK Schmidt Telescope in Australia. Various line ratios, discontinuities, and continuum intensities have been used as criteria. The scheme is a 2-D system (spectral type — luminosity class), and the criteria are given in graphical form. The typical accuracy of our scheme is 2 subdivisions of spectral type. Comparison of our types with those given by other authors for common stars yields a very good correlation with deviations within our accuracy.

The present celestial reference frame is represented by the positions and proper motions of the 4500 stars in the Fifth Fundamental Catalogue (FK5). For practical applications it is extended by the catalogues IRS (International Reference Stars) and PPM (Positions and Proper Motions). According to resolutions of the IAU the FK5 system will be replaced by a purely kinematically defined reference system defined by extragalactic radio sources in the near future. At optical wavelengths, the forthcoming Hipparcos Output Catalogue will be linked to the extragalactic system.

The methods used to assess the quality and the system of the AGK3U (Bucciarelli et al. 1992) are reviewed. However, the point of this paper is that all star catalogs should undergo a comparable evaluation. It is too easy to follow the recipes of statistics, examine internal measures, declare success, and move on. All recipes can be ruined with the wrong ingredients and the theorems of statistics are no different. Since it is impossible to verify that the hypotheses are a priori fulfilled, one has to be resourceful and devise methods which fully test the outcome a posteriori. The AGK3U has been so examined; our efforts are described herein. (The AGK3U is the first attempt to combine astrometrically reduced Schmidt plate positions with those from astrographic plates [Taff 1989; Taff et al. 1990]. Our sub-plate code will be available soon because of NASA support to make it portable.)

The original version of the HST Guide Star Catalog (GSC 1.0, Lasker et al. 1990, and following papers) contains systematic errors which typically begin from the plate center and increase with radius (Taff et al. 1990a). In a residual plot against CAMC stars this shows as high correlations between neighboring residuals. This is due to the inability of a global polynomial to map the metric of a Schmidt plate. A new reduction of the GSC measures (GSC 1.2) based on the subplate method (Taff 1989) is partially complete. We report on it for the northern data from the Palomar Quick V plates, with 20 minute exposures, nearly all taken in the 1982–1984 interval.

Schmidt plate-based surveys of the sky are currently the largest sources of positions and colours of stellar and non-stellar objects (see e.g. Lasker 1993). This situation is likely to remain such through the next decade when new and improved CCDs will probably replace the photographic medium. But even then deep photographic surveys will constitute important first epoch material for the derivation of better proper motions. Accordingly, we should do our best to improve astrometry and photometry of Schmidt plates.