An New Adaptive Quantization To Reduce Blocking Effect

In this paper, we present a method to reduce the blocking effects in block based coding schemes. Based on our observation, the blocking effect appears mainly in the region with fast or complex motion which is not well compensated. But this region tends to be coarsely quantized in the traditional method using only the spatial activity measure. [2] In order to quantize this region finely enuogh, we introduce a new activity measure, the slope activity. By using this measure together with the spatial activity measure, we perform the adpative quantization for each MB in a picture. Our approach shows superior picture quality compared to the MPEG TM5. Particularly in fast or complex motion scenes, the blocking effects in the object boundaries are much suppressed than that of MPEG TM5, therefore generating much smoother image for comfortable viewing. Also, our method shows higher PSNR than that of the MPEG TM5. I . INTRODUCTION The MPEG standard provides a fine method which can compress moving pictures with low bitrates [l], [2]. But due to its lossy characteristics, distortions are highly visible if not enough bits are used. Many different types of distortions are inevitable if this kind of compression techinique is used [4]. Above all, the blocking effect is known to be the major artifact in a block based coding scheme. Based on our observations of video sequences, blocking effect appears mainly in the regions having nontranslational, fast, or random motion. The reason is that motion compensation itself is done roughly in these regions leaving lots of disparities along the MB(Macro Block) boundaries in the prediction error image. And in traditional approaches, many MBs in these regions tend to be quantized coarsely by using only the spatial activity [2], [3]. So the disparities in the MB boundaries are not well compensated. And if these regions exist in successive frames in a video sequence, blocking artifacts are quite noticeable. To resolve this limitation of the spatial activity measure, we propose a new activity measure, a slope activity measure, by which we can find the region with the complex motion. Using this measure together with the spatial activity measure, we perform the adaptive quatization of each MB in a picture. 11. DETECTION OF THE BLOCKY REGION In the previous section, we pointed out that more blocking effects are generated in the regions where the motion is not well compensated. Fig. l.(a) shows a difference image of the football sequence. We can see that the lots of disparities in the MB boundaries mainly appear in the regions with complex motion. These disparities are similar to blockiness, in order to detect these region, we adopt Mean Square Difference of Slope(MSDS) which detects the degree of blockiness on the boundaries by measuring the gradient between the neighboring blocks along the boundary [5]. We apply this MSDS criterion with some modification to difference image and detect the not well motion compensated regions. Fig. l.(b) shows that the slope of MB is high in the region with the complex motion. Here, large grey level value is assigned to the MB which has higher slope activities. 111. ADAPTIVE QUANTIZATION The regions with low spatial activities are already quantized finely in MPEG TM5, so we can expect that the blockiness is not much visible in those regions. Whereas, in the regions with large spatial activities, we apply MB’s slope information. The slope activity, P-actj, is defined as follows, slpj + 2 x avg-slp 2 x slpj + avg-slp P-act . Where slpj is the j t h MB’s slope in current picture and avg-slp is the average slope of the previous same type picture. Then, the new activity measure, N-actj, consisting of two activity measures is as follows, N-actj = a ( S a c t j ) x S-actj + (1 -a(S-actj)) x P-actj (2) where the spatial activity measure in TM5 is redefined as S-actj. The weighting, a( . ) , being used in eq.(2) is the function of S-actj shown in Fig. 2. The two thresholds indicate the transition region between two activity measures. The empirically obtained threshold values, SO and S1, are set to 0.9 and 1.3 respectively. By using this new activity measure, we efficiently eliminate the blocking effects in a picture by quantizing MBs adaptively. IV. EXPERIMENTAL RESULTS From our experiments, we have found that the proposed scheme produces better subjective quality and an average of 0.4 0.5 dB increase in PSNR over that of MPEG2