A note on grey level-intensity transformation: effect on HVS thresholding

Ab.l/fllel: An attempt is made in this note to investigate the effeet on edge extraction when the theory of HVS (human visual system) based thresholding [ll is made to operatc on the mtensity domain ofa grey tone image. A family of 5 and S-[ functions is considered to take eare of the grey level to intensity transformation. The performance of the system is also quantitatively ana­ lysed using the measure 'entropy' of a fuzzy set. An algorithm for thresholding based on the facts of human psychovisual phenomena or the response of the human visual system (HVS) was reported in [1]. We call this method HVS based thresholding throughout this text. This made it possible to select automatically (without human intervention) the thresholds for detecting the significant edges as per­ ceived by human beings. The threshold value adapts the background intensity according to the criterion governed by a characteristic of one of the De Vries-Rose, Weber and saturation regions. These regions are characterised in the intensity do­ main by (l a) .JB T ce B, and (lb) .JB T ce B 2 , (1 c) respectively. .J H T and B denote the incremental threshold and background intensity. Equation (I) respresents HVS characteristics. The algorithms were found for a wide range of images to provide a satisfactory improvement in the performance in the conventional edge detection process. It is to be noted that while implementing the above mentioned algorithms for a digital grey tone image, it was implicitly assumed that there exists a linear relation between the discrete intensity and grey level so that the intensity values can well be substituted by its corresponding transformed grey level values without affecting the above characteris­ tic equations (1) of the HVS. In practice, when scanning an image, there is a non-trivial transformation function which links in­ tensities going in and grey scales coming out. The question may therefore arise (in fact, raised by many readers) how the results [IJ will be affected if one considers the exact transformation used during the digitisation process. A typical form of the response C = fCB) between the digitised grey value (C) and the intensity value (B) (sensor input luminance value) is shown in Figures I [2]. The compensation methods to have a distortion free grey image are discussed in the lit­ erature [2J. An attempt is made in this letter to investigate the effect on the …