VISAFF: Speaker-Centered Visual Affective Feature Learning for Emotion Recognition in Conversation