Cloud OCR SDK uses the same core as ABBYY FineReader and that core is currently not very good at recognizing text on photos as the example picture you linked to. We're actively working on improving our OCR internal engine and once that work is complete text on photos recognition will be greatly improved.
The picture you provide as the example has text made with distinct color. You might try a preprocessing filter that would eliminate all but the text - for example, make areas of that color black and the rest of the image white. That would greatly increase chances for successful OCR of such images. Other than specialized image preprocessing there's nothing that can be done at the moment.