Related Work

Most of the literature deals with segmentation based on either colour or texture. Although colour is an intrinsic attribute of an image and provides more information than a single intensity value, there has been few attempts to incorporate chrominance information into textural features [4]. This extension to colour texture segmentation was originated by the intuition that using information provided by both features, one should be able to obtain more robust and meaningful results.

A rather limited number of systems use combined information of colour and texture, and even when they do so, both aspects are mostly dealt with using separate methods [7]. Generally, two segmentations are computed for colour and texture features independently, and obtained segmentations are then merged into a single colour texture segmentation result with the aim of preserving the strength of each modality: smooth regions and accurate boundaries using texture and colour segmentation, respectively [8,9]. The main drawback is related to the selection rule for assigning the appropriate segmentation labels to the final segmentation result, where segmentation maps disagree with each other.

It is only recently that attempts are being made to combine both aspects in a single method. Three alternatives to feature extraction for colour texture analysis appear to be most often used and they consist of: (1) processing each colour band separately by applying grey level texture analysis techniques [10,11], (2) deriving textural information from luminance plane along with pure chrominance features [12,13], and (3) deriving textural information from chromatic bands extracting correlation information across different bands [14,15,16,17].

Our proposal can be classified in the second approach, considering chromatic properties and texture features from the luminance, which facilitates a clear separation between colour and texture features.