Truncated gray-coded bit-plane matching based motion estimation and its hardware architecture

Abstract

Özetçe Bu çalÕúmada özellikle düúük güçlü mobil cihazlar için uygun, düúük bit gösterimi temelli bir hareket kestirimi yaklaúÕmÕ ve donanÕm mimarisi önerilmiútir. Önerilen yöntemde hareket kestirimi iúlemi, gray kodlanmÕú piksellerin bit düzlemlerinden, belirli sayÕdaki en de÷erliksiz düzlemin kesilmesi ve kalan bit düzlemlerinin uyumlanmasÕ ile gerçekleútirilmiútir. Deneysel sonuçlar önerilen yöntemin di÷er bir-kesme temelli yöntemlere göre daha üstün bir hareket kestirimi baúarÕmÕna sahip oldu÷unu göstermiútir. Tasarlanan donanÕm mimarisi di÷er 8-bit/piksel temelli mimarilere göre çok küçük bir yonga alanÕna ve düúük güç tüketimine sahip oldu÷u için güncel kodlayÕcÕlarla kolayca tümleútirilebilir. AbstractIn this paper, an efficient low bit-depth representation based motion estimation approach which is particularly suitable for low-power mobile devices is proposed. Motion estimation is carried out using bit truncated gray-coded image pixels in the proposed approach. The hardware architecture of the proposed motion estimation method is also designed to show the effectiveness of the proposed approach. It is shown that the proposed approach provides improved motion estimation accuracy compared to the other bit-truncation based approaches. The proposed hardware architecture has low hardware complexity and consumes very low power compared to the 8-bits/pixel based hardware architectures thus, it can be easily integrated to the state of the art video encoders.978-1-4244-4436-6/09/$25.00 ©2009 IEEE 978-1-4244-4436-6/09/$25.00 ©2009 IEEE

Adaptive bit-plane selection-based low complexity motion estimation for screen content coding

“…In the matching stage, the proposed method has comparable computational complexity with the ME methods with C-1BT [14] and SGCME [16] and significantly lower complexity compared to T-GCBPM [15]. In general, when the ME performance of the proposed method given in Table 2 is assessed together with the computational complexity given in Table 4, the efficiency of the proposed method becomes clear.…”

“…The PSNR results of the 1BT [12], C-1BT [14], T-GCBPM [15], SGCME [16], and FBMESC [17] methods together with the proposed method are given in Table 2. As seen from this table, the proposed method is not only able to outperform existing 1-bit depth representation-based methods such as 1BT [12], SGCME [16], and FBMESC [17] but also two-bit depth representation-based methods such as C-1BT [4] and T-GCBPM [15] (NTB = 6). It is also important to note that the proposed method provides similar ME accuracy compared to the NTB = 5 case where three separate bit-planes are used for matching at significantly higher complexity compared to the proposed method.…”

“…In [15], the T-GCBPM approach was presented. After converting binary values to Gray-coded values via simple EX-OR based operations a truncation is performed to use a few of the most significant bits.…”

“…In this work, we aim to benefit from strong parts of the methods presented in [16] and [17] and improve them in a novel way. The advantage of using Gray-coded bit-planes compared to natural binarily coded ones was discussed in [15,16,19]. In general, Gray codes have a single digit change in successive pixel intensity values and thus they are more suitable for matching computation.…”

“…As seen from Figure 4a and Figure 4e, the most significant bit-planes may not contain reliable information for matching by themselves, especially for the flat image regions where neighbor pixels have the same binary values and thus there is not any discriminating structure for block matching in higher bit-planes, whereas lower bit-planes may include some noise like binarization effect in certain regions, as in Figure 4d. However, as shown in [15], the most significant four bit-planes generally provide enough information for form matching. In this paper, we propose to select the Gray-coded bit-planes to be included in matching criterion computation by simply checking the edge map at the most significant bit-plane only.…”