Both display and file are quite slow, and have the potential to bring even quite capable systems to their knees dealing with many multiple files. A small test:

     $ du -h *.png --total | tail -n 1
     9.2M    total

     $ ls -l *.png | wc -l
     107

     $ /usr/bin/time file *.png
-->  0.37user 0.26system 0:06.93elapsed 9%CPU (0avgtext+0avgdata 37232maxresident)k
     22624inputs+0outputs (9major+2883minor)pagefaults 0swaps

     $ /usr/bin/time identify *.png
-->  0.56user 0.22system 0:06.77elapsed 11%CPU (0avgtext+0avgdata 25648maxresident)k
     34256inputs+0outputs (119major+2115minor)pagefaults 0swaps

By reading only the bytes necessary, this operation can be significantly sped up.

     $ /usr/bin/time ./pngsize *.png
-->  0.00user 0.00system 0:00.03elapsed 12%CPU (0avgtext+0avgdata 1904maxresident)k
     0inputs+0outputs (0major+160minor)pagefaults 0swaps

Here is pngsize:

#include <stdio.h>
#include <fcntl.h>
#include <unistd.h>
#include <arpa/inet.h>
#include <err.h>
#define oops(syscall) { printf("error processing %s: ", argv[i]); \
        fflush(0); perror(syscall"()"); continue; }
int main(int argc, char **argv) {
    int fd, i;
    uint32_t h, w;
    if (argc < 2) { printf("%s <pngfile> [pngfile ...]\n", argv[0]); exit(0); }
    for (i = 1; i < argc; i++) {
        if (argc > 2) printf("%s: ", argv[i]);
        if ((fd = open(argv[i], O_RDONLY)) == -1) oops("open");
        if (lseek(fd, 16, SEEK_SET) == -1) oops("lseek");
        if (read(fd, &w, 4) < 1) oops("read");
        if (read(fd, &h, 4) < 1) oops("read");
        printf("%dx%d\n", htonl(w), htonl(h));
        if (close(fd) == -1) oops("close");
    }
    return 0;
}

This method is much faster than using a library which loads the PNG forwards, backwards and sideways just to get the image size :P (Consider the code carefully before feeding it a directory full of arbitrary PNGs of course.)

The code uses inet.h for htonl() to de-endian-ize the header byte ordering.