2 Observations and data reduction

Observations of 12 BCDGs were conducted at ESO-La Silla on the 2.2 m telescope, with the infrared camera IRAC2 a $256\times 256$ pixel NICMOS chip. We used the focal configuration giving 0.507arcsec/pixel with a field of view of $2\times2$ arcmin. Table 1 summarizes of the observations.

Column 1: Name of the object.
Column 2: Apparent B magnitude from PaperII.
Column 3: Photometric classification in the B band coded as follows: d0 have pure or almost pure exponential light profiles, cd have composite profiles with dominant exponential. e0 have pure or almost pure r^1/4law profiles, ce have composite profiles with dominant r^1/4.
Column 4: Observing date.
Column 5: Total exposure time in minutes in J, H and $K_{\rm s}$bands respectively.
Column 6: S/N, signal to noise ratio at isophotal level K = 22.5 mag arcsec^-2, in $K_{\rm s}$ band.

In the near infrared, we did not expect the observed size of our galaxies to exceed half the size of view the detector, therefore, we used the in-field chopping technique. The BCDG in our sample with the largest apparent diameter, has a radius of 60'' at B = 27 mag arcsec^-2. If only an evolved stellar population dominates at this radius, then $K \sim$  23 mag arcsec^-2 at $r\sim60''$, which is below the detection limit of a ground-based 2 m class telescope. We obtained several mosaics for each galaxy in each band ( $JHK_{\rm s}$) with offsets of about 20'' in the East-West or North-South directions. The in-field chopping method is very effective because no time is lost for sky observations, thus allowing all of the time to be spent on the target. It also provides a better sky background subtraction because the sky background determination is made at the same position as the object and almost simultaneously. The observational strategy was to complete the observations for a given galaxy at least in a given filter per night, thus we did not have to combine images taken during different nights. Sequences of internal flat-fields in each band and bias exposures were obtained in the afternoon and in the morning, and dark exposures at the end of each night.

Each elementary frame was checked against all others within a given mosaic, and frames with large sky variations were rejected before combining the frames. The quality of the sky background varies from one mosaic to another because the weather conditions were not very stable. Moreover, for the combination of the frames, we used a strict rejection algorithm (minmax algorithm with high-rejection) to eliminate shadows of compact objects, like field stars, that remained on the sky background frames and would otherwise perturb the signal in the sky subtracted images. The sky subtraction procedure also accounts for bias subtraction and dark current removal. For flatfielding, we used the internal flatfields obtained before or after each observing night.

The biases were also used to produce a map of bad pixels, which is taken into account during the final combination of the frames. Since our strategy was to observe one band after the other, we could combine all frames at once, in each band for each galaxy.

Surface photometry was done using "user-built'' procedures in MIDAS consistent with those used in the visible (see PaperI for a detailed explanation). Magnitude calibrations were performed using the standard stars from the Carter system observed at the same zenith distances as the galaxies. Since we observed the galaxies in the $K_{\rm s}$ band, we used the correspondence equations provided by the 2.2 m telescope Team (technical report IV: Lidman & Storm 1995) to transform the $K_{\rm s}$ magnitudes into the Carter K magnitudes.