Our MIR observations were obtained using ISOCAM, a $32\times32$ pixel array (Cesarsky et al. 1996a) on board the ISO satellite (Kessler et al. 1996). Each system was observed with broad band filters ranging from 5 to 18 $\mu$ m in a $2\times2$ raster with 6 pixel offsets and a lens producing a pixel field of view (PFOV) of 1.5 $^{\prime\prime}$ , resulting in a final image of $57\hbox{$^{\prime\prime}$ }\times57\hbox{$^{\prime\prime}$ }$ . This enabled us to obtain images with a spatial resolution of 3 $^{\prime\prime}$ (at 6 $\mu$ m) to 4.5 $^{\prime\prime}$ (15 $\mu$ m) limited by the pixel size at 6 $\mu$ m and by the full width at half maximum (FWHM) of the point spread function (PSF) at 15 $\mu$ m. We note the ISOCAM filters by their name and central wavelength. The wavelength range in $\mu$ m covered by each filter was: LW2 (5.0-8.5), LW3 (12.0-18.0), LW4 (5.5-6.5), LW6 (7.0-8.5), LW7 (8.5-10.7), LW8 (10.7-12.0), LW9 (14.0-16.0). At subsequent sections in this paper we will refer to the measured flux densities using the various filters as $f_{x~\mu\rm m}$ where x is the central wavelength of each filter in microns.

Spectrophotometric observations were also obtained with the circular variable filter (CVF) for IRAS 23128-5919, the brightest of our sources. The CVF covers a spectral range from 5 to 16.5 $\mu$ m with a 1.5 $^{\prime\prime}$ PFOV and a spectral resolution of 50. Each integration step was composed of 12 images with 5.04 s integration time and during the CVF scan the wavelength step varied between 0.05 and 1.11 $\mu$ m. Details on the observing parameters are summarized in Table 2.

The data were analyzed with the CAM Interactive Analysis software (CIA). A dark model taking into account the observing time parameters was subtracted. Cosmic ray contamination was removed by applying a wavelet transform method (Starck et al. 1997). Corrections of detector memory effects were done applying the Fouks-Schubert's method (Coulais & Abergel 2000). The flat field correction was performed using the library of calibration data. Finally, individual exposures were combined using shift techniques in order to correct the effect of jittering due to the satellite motions (amplitude $\sim$ 1 $^{\prime\prime}$ ). A deconvolution using multiscale resolution techniques (Starck et al. 1999) was subsequently applied to estimate the physical size of the quasi-point like sources responsible for the infrared emission in our data (see Sect. 3.1).

The details of the analysis of the ISOPHOT-S data of the three galaxies, which we also include in this paper for reasons of comparison, are published by Rigopoulou et al. (1999).

Based on three different observations of IRAS 19254-7245 taken with identical LW filters but with different roll angle, integration times per exposure (2 s and 5 s) and PFOVs, as well as on similar analysis of other ISOCAM-CVF and ISPHOT-S observations, we estimate that the uncertainty of our photometry measurements is $\sim$ 20 $\%$ (see Table 3).

**Table 1:** Properties of the three IRAS systems.
Target	RA	DEC	z	$F_{\nu}$ (Jy)				$D_{\rm L}$	$\log(L_{\rm FIR}$ )	$\log(L_{\rm IR}$ )
IRAS Name	J2000.0	J2000.0		12 $\mu$ m	25 $\mu$ m	60 $\mu$ m	100 $\mu$ m	(Mpc)	( $L_{\odot}$ )	( $L_{\odot}$ )
19254-7245	19 $.\!\!^{\rm h}$ 31 $.\!\!^{\rm m}$ 21.6 $.\!\!^{\rm s}$	$-72\hbox{$.\!\!^\circ$ }39\farcm20.8\hbox{$.\!\!^{\prime\prime}$ }$	0.0617	0.22	1.24	5.48	5.79	250	11.68	12.01
23128-5919	23 $.\!\!^{\rm h}$ 15 $.\!\!^{\rm m}$ 46.9 $.\!\!^{\rm s}$	$-59\hbox{$.\!\!^\circ$ }03\farcm14.2\hbox{$.\!\!^{\prime\prime}$ }$	0.0446	0.24	1.59	10.80	10.99	180	11.69	11.96
14348-1447	14 $.\!\!^{\rm h}$ 37 $.\!\!^{\rm m}$ 38.2 $.\!\!^{\rm s}$	$-15\hbox{$.\!\!^\circ$ }00\farcm23.9\hbox{$.\!\!^{\prime\prime}$ }$	0.0823	<0.14	0.49	6.87	7.07	335	12.05	12.27

Table note: The far-infrared and infrared luminosities are calculated using $L_{\rm FIR}=3.94\times10^5\times D$ (Mpc) $^2(2.58 \times{f}_{60}+f_{100}$ ) and $L_{\rm IR}=5.62\times10^5\times D$ (Mpc)²( $13.48\times f_{12}+5.16\times f_{25}+2.58\times f_{60}+f_{100}$ ) respectively, where the luminosity distance is defined as $D_{\rm L}= \frac{c}{H_0q_0^2}(zq_0+(q_0-1)(\sqrt{(1+2q_0z)}-1)$ (see Sanders & Mirabel 1996).

**Table 2:** ISOCAM observing parameters.
Target	ISOCAM Filter:	LW2	LW3	LW4	LW6	LW7	LW8	LW9	CVF
	Filter Center:	6.75 $\mu$ m	15 $\mu$ m	6 $\mu$ m	7.75 $\mu$ m	9.62 $\mu$ m	11.4 $\mu$ m	15 $\mu$ m	-
IRAS 19254-7245¹		15.3	15.3	15.4	15.2	15.4	15.3	15.4	-
IRAS 19254-7245²		7.1	7.0	11.3	-	8.3	-	-	-
IRAS 19254-7245³		3.4	3.6	-	-	-	-	-	-
IRAS 23128-5919⁴		7.2	7.0	11.5	-	8.5	-	-	-
IRAS 23128-5919⁵		-	-	-	-	-	-	-	148.7
IRAS 14348-1447⁶		8.6	8.4	-	-	-	-	-	-

Table note: The numbers following each galaxy denote the total on-source exposure time (in minutes) for each filter used, and two galaxies were observed more than once under different configurations, the details of which are: (1) IRAS 19254-7245 observed in proposal CAMACTI2 (PI I.F. Mirabel), 7 LW filters, integration time per frame Tint = 5s, pfov = 1.5 $^{\prime\prime}$ . (2) IRAS19254-7245 observed in proposal CAMACTIV (PI I.F. Mirabel), 4 LWfilters, Tint = 2s, pfov = 1.5 $^{\prime\prime}$ . (3) IRAS19254-7245 observed in proposal SAM12N_2 proposal (PI L. Spinoglio), 2 LW filters, Tint = 2s, pfov = 3 $^{\prime\prime}$ . (4) IRAS 23128-5919 observed in proposal CAMACTIV (PI I.F. Mirabel), 4 LWfilters, Tint = 2s, pfov = 1.5 $^{\prime\prime}$ . (5) IRAS 23128-5919 observed in proposal CAMACTI2 (PI I.F. Mirabel), CVF, Tint = 5s, pfov = 1.5 $^{\prime\prime}$ . (6) IRAS 14348-1447 observed in proposal CAMACTIV (PI I.F. Mirabel), 2 LW filters, Tint = 2s, pfov = 1.5 $^{\prime\prime}$ .

2 Observations and data reduction