Abstract: A compressor for compressing a block of feature vectors representing a feature associated with image elements, includes means (20, 22) for determining the distribution of the feature vectors, means (20, 24, 26, 28) for transforming each point pattern in a predetermined set of point patterns to fit the determined distribution and a selector (30) for selecting a transformed point pattern that best fits the determined distribution. Furthermore, an encoder (32) represents the block of feature vectors by an identifier identifying the selected point pattern in the set of point patterns, parameters representing the transformation associated with the selected point pattern, and an index for each feature vector representing the nearest point in the transformed selected point pattern.