Prominent Boundaries and Foreground Detection Based Technique for Human Face Extraction from Color Images Containing Complex Background

The paper proposes a novel method for human face extraction from color images, characterized by uncontrolled illumination conditions and complex backgrounds. The method incorporates stationary Haar wavelet transform & proximity influence for prominent boundaries detection and watershed transform, proximity influence & morphological operations to separate foreground / background along with region and color attributes for human face extraction. The method exploits redundancy by coalescing local color cues of all color channels to emphasize reliable processing to precisely detect the human face by avoiding under-segmentation and reducing over-segmentation & artifacts. The method has been tested on face-image collection of standard database and on images captured by an amateur photographer for various complex backgrounds having diversified textures, varied illumination conditions and multiple background objects. The presented results show the effectiveness of the method for human face extraction, proving it suitable as an input to applications like digital album catalogue, content based image retrieval, face recognition and facial expression recognition.