If you want to save data that is derived from a Javascript canvas.toDataURL() function, you have to convert blanks into plusses. If you do not do that, the decoded data is corrupted:

<?php
  $encodedData = str_replace(' ','+',$encodedData);
  $decocedData = base64_decode($encodedData);
?>

Base64 for URL parameters/filenames, that adhere to RFC 4648.
Defaults to dropping the padding on encode since it's not required for decoding, and keeps the URL free of % encodings.

<?php
function base64url_encode($data, $pad = null) {
    $data = str_replace(array('+', '/'), array('-', '_'), base64_encode($data));
    if (!$pad) {
        $data = rtrim($data, '=');
    }
    return $data;
}
function base64url_decode($data) {
    return base64_decode(str_replace(array('-', '_'), array('+', '/'), $data));
}

I had some trouble trying to let base64_decode decode base64-strings longer than ~5k chars.

The base64-decoding function is a homomorphism between modulo 4 and modulo 3-length segmented strings. That motivates a divide and conquer approach: Split the encoded string into substrings counting modulo 4 chars, then decode each substring and concatenate all of them.

Then instead of 

<?php $decoded = base64_decode($encoded); ?>

for big $encoded strings, it's saver to use

<?php
$decoded = "";
for ($i=0; $i < ceil(strlen($encoded)/256); $i++)
   $decoded = $decoded . base64_decode(substr($encoded,$i*256,256));
?>

where 256 can be replaced by a sufficiently small modulo 4 natural.

This function supports "base64url" as described in Section 5 of RFC 4648, "Base 64 Encoding with URL and Filename Safe Alphabet"

    <?php
    function base64url_decode($base64url)
    {
        $base64 = strtr($base64url, '-_', '+/');
        $plainText = base64_decode($base64);
        return ($plainText);
    }
    ?>